A NNALES DE L’ INSTITUT F OURIER

B JÖRN E. J. DAHLBERG
C ARLOS E. K ENIG
J ILL P IPHER
G. C. V ERCHOTA
Area integral estimates for higher order
elliptic equations and systems
Annales de l’institut Fourier, tome 47, no 5 (1997), p. 1425-1461
<http://www.numdam.org/item?id=AIF_1997__47_5_1425_0>

© Annales de l’institut Fourier, 1997, tous droits réservés.

L’accès aux archives de la revue « Annales de l’institut Fourier »
(http://annalif.ujf-grenoble.fr/) implique l’accord avec les conditions gé-
nérales d’utilisation (http://www.numdam.org/conditions). Toute utilisa-
tion commerciale ou impression systématique est constitutive d’une in-
fraction pénale. Toute copie ou impression de ce fichier doit conte-
nir la présente mention de copyright.

Article numérisé dans le cadre du programme

Numérisation de documents anciens mathématiques
http://www.numdam.org/
Ann. Inst. Fourier, Grenoble
47, 5 (1997), 1425-1461

AREA INTEGRAL ESTIMATES FOR HIGHER

ORDER ELLIPTIC EQUATIONS AND SYSTEMS

by B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER

and G.C. VERCHOTA

Introduction.

The classical formulation of boundary value problems for constant

coefficient elliptic operators, or systems of operators, involves continuous
data on the boundary of a (smooth) domain and leads to the existence and
uniqueness of solutions continuous up to the boundary. If the given data is
not continuous but exists only in some L^ space, it may still be possible to
obtain unique solutions with this data. The data would be taken on in the
sense of nontangential limits and one obtains nontangential L^ estimates
on the solution which guarantee uniqueness in this class.
Let us recall the situation for harmonic functions in, say, the upper
half space R^ [30]. Given data f(x) € L^R71"1), 1 < p < oo, the
function u(x^y) = Py * f(x) (where Py(x) denotes the Poisson kernel)
is the unique solution to the Dirichlet problem : AIA = 0, u y=Q = f{x)
with appropriate decay at oo, and for which the nontangial maximal
function of u is also in L^R77'"1). The nontangential maximal function
of u is Nu{x) = sup{|zi(.r',^)| : \x' - x\ < q/}, the supremum of
values of u taken in the cone T(x) = {{x'\y) : \x' — x\ < cy} with
aperture determined by c. The nontangential maximal function plays a

The last three authors supported in part by the N.S.F.

Key words'. Elliptic systems - Lipschitz domains - Area integral.
Math. classification: 35J55 - 35J40 - 42B25.
1426 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

rule analogous to the role of the Hardy-Littlewood maximal function in the

context of the Lebesgue differentiation theorem - via this maximal function,
nontangential limits are controlled. (In fact, for the Poisson integral of an
L^ function u(x^ y) = Py * f(x)^ Nu(x) and the Hardy-Littlewood maximal
function of / are comparable.) In the setting of harmonic functions, the
Z/^-Dirichlet problem is uniquely solvable if, for every / e L^R71"1),
there exists a harmonic u which converges (nontangentially, i.e. through
sequences (x1', y) —>• (.r, 0) restricted to the cone at x) to the data, and
which is in the class Nu € L33. Then as a consequence of linearity one has
the estimate ||7V(u)||^p(^n-i) < C||/||^p(^n-i).

The L^-behavior of the nontangential maximal function gives very

precise control over the growth of solutions to this elliptic boundary value
problem, but it is not the only means of obtaining such control. One can
also measure the P3 norm of the square function of solutions ZA, denned for
x e V1-1 by S(u)(x) = {Jp^ IVn^'.^IV-'Wch/}^. There are several
reasons one might prefer L13 estimates on the square function of a solution :
the 'geometric5 content of this quadratic expression, the connection with
measuring Sobolev/Besov norms and the invariance of the square function
under certain important linear operators that may not always be singular
integral operators (e.g. Riesz transforms).

Our objective in this paper is to demonstrate the L^ equivalence

(0 < p < oo) between the nontangential maximum function and the square
function of solutions to the homogeneous equation for higher order elliptic
systems on Lipschitz domains. Such results, in special cases, have been
proven in earlier works [10], [25], [5]. Here we make no restriction on the
order of differentiation or the size of the (determined) systems. The systems
considered consist of only a principle part, satisfy the Legendre-Hadamard
condition and are real symmetric.

In extending the square function estimates from the situation of

harmonic functions in M^:, there are three essential difficulties to be
overcome. First, the boundary of our domain f2 is not smooth and the
quantity which replaces the factor of y = dist((a*, ^/); R11'"1) in the definition,
namely S(X) = dist(X,c%"2), is not more than once differentiable. Second,
the elliptic operators are of higher order and may not have a quadratic
form which is coercive (see [2] for example). In particular this means that
the quadratic form associated to the operator is not in an obvious way
related to the quadratic expression used in the definition of the square
function. (Such relations exists when the operator can be written as a sum
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1427

of squares however.) Finally, the situation of higher order systems is yet

more complicated, and we develop an argument for reducing the case of
systems of equations to single equations.
We can illustrate the elementary new arguments used to prove the
square function estimates on nonsmooth domains just in the case of
functions harmonic in a Lipschitz domain ^2. The idea is to use a variant
of the 'adapted' distance function of Dahlberg [11] invented by C. Kenig
and E. Stein [2]. The adapted distance function has been used to prove
quadratic estimates in other settings - [22] for Clifford valued monogenic
functions and [18] for solutions to parabolic equations.
Let Q C R71 be the domain above the graph of a Lipschitz function
and let 6(X) for X e ^2 be the adapted distance function as discussed in
§1 below. The properties of 6 we need here are (i) 6(X) w dist(X, 9^), (ii)
——W = Dn6(X) > c' > 0, (iii) |V^(X)| < c", (iv) ^IVV^X)!2^
O^n
is a Carleson measure, and (v) VDy^ = (Dn6)2 Y, VZ^j where each 6j
3
has the property that \Dn6j(X)\26~l{X)dX is a Carleson measure and
\D^W\<c".
Let us now give the argument which proves

( S2{u)d(TW ( N2(u)da
JQQ. Jao,
for u harmonic in f^, as above, and which illustrates the main ideas
needed to prove the analogous result for higher order homogeneous elliptic
equations possessing coercive bilinear forms.
We first note that to dominate the nontangential maximal function
of a harmonic function u in ^2 by the square function of n, it suffices to
dominate |H|L2(^,d<r) by \\S(u)\\^^,da)- For harmonic functions, this
follows from Dahlberg's theorem on the L2-solvability of the Dirichlet
problem ([9]). For biharmonic and polyharmonic functions, one needs the
L2 solvability results of [14] and [32], and for the general case of higher
order elliptic equations or systems of equations, one invokes the results of
[27] and [33]. With this in mind, let us take a function n, harmonic in
^ C R71, and show that

I u2d(T<C ( S2{u)da=C [ 6(X)\\7u(X)\2 dX

Jan JQQ Jxw
where 6(X) is the adapted distance function. If N = (TVi,..., Nn) denotes
the unit normal vector to the boundary of f^, then Nn > c > 0 for some c
 1428 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER &: G.C. VERCHOTA

that depends only on the Lipschitz constant of 0. Thus,

U2da c U2Nnda D dx
L =^2LUDnudx=-l2 ^uD dx
-L =- i ^
'-^(^^-^(t^-
The first integral above is zero, since 6 = 0 on 9fl, and the second integral
above gives rise to three terms, upon distributing the derivative

=
/^(^
z ^ ^ ^ /„ ^"^. /„ <"M^) ^
=1+2+3.

Term 3 is bounded by c(f^6\D^)12 . (/^|VV^|2)5, since D^6 >

d > 0. By the Carleson measure property of 6\^6\2dX, this is in turn

controlled by (/^|V^x)' . (f^N^n)da)12. Term 1 is equivalent to

fan 's'2(^()c(o" and it remains to handle term 2. Here we shall introduce the
w
quantity 1 = ^ N^, where at each X N = and so TV, = -D^. Hence

2=
/^^"^^^
J

=
^ ^ [^AAU + ^ JY,-£»n^uJV,] dX.
J=l
Now N.DnD^u = (A^,^ - N^D^u + M^,A,u. The expression
Aj^n - NnDj for 1 < j ^ n - 1 is a tangential derivative to the level
sets of 6, and we shall integrate by parts noting that tangential derivatives
of 6 are zero. This may be done by using the co-area formula So for
1 < J ^ n - 1,
r <^v".
/ ———(NjD^-N^Dj)DnudX
Jfl, -^n^

=
r00.,, f uN,
k ^l^WwW--^^^
=f^nWN^-NM(^^X
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1429

where dot denotes surface measure along S = t. It is by this application of

the co-area formula that the integration by parts in the proof of the main
lemma of §2 is done. When the tangential derivative falls on u, we get a
term equivalent to f^6\^u\2 and when the derivative falls on {Dn6)~1 or
|V^|~1 we again use the Carleson measure property and Cauchy-Schwartz
to get a 'mixed5 term of the form ( f^ N2^)) ^ ' ( f^ S2(u)) ^. The term
where the derivative falls on Nj is also controlled by a product of this
/7?7
form. For 1 < j < n, it remains to evaluate f^ ——NjNnDjDnU. Again,
Dn0
we introduce tangential derivatives by expressing NjDnDju = [NjDn —
n—l g^,
NnDju]DjU + N^DjDjU. The expression ^ ——N^DjDjU cancels with
j=l L/u0

——N^DnDnU
0
by using harmonicity of u and the term with the tangential
L/n
derivatives is handled exactly as before, integrating by parts. This settles
one half of the L2 equivalence. See Theorem 1 of §2 below.
We now give the argument for the converse inequality, which will
require using the additional Carleson measure property (v) above
/* n /*
2 / y|D,u(X)|2 6(X)dX = \ A(n2)^
^ ^i Jn

(1) = / div[W(^2)] - VW(n 2 )

Jfl
=- [ div(^V6) + u2^
Jfl.
== { u2\^6\da+ [ u2^.
JQ^. J^i
The first integral above is bounded by the desired term f N2(u)do•
and to handle the second integral, we introduce the quantity —r— and
Dn6
integrate by parts using Dn. Thus

/ u2^^^ = - I SD^u2^——] dX,

JQ ^nO Jo. L Dn6\
and there are three terms arising from distributing the differentiation:
I = - f 6 . (2uDnu)^6(Dn6)~1
Jo.
^^[^^.(fnW——)32
-•Jfl ' ^Jsi Onb/

^\\S(U)\\L.W\\N(U}\\LW.
1430 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

TT f f. 2^DnDn6
11= / 6U ^6 ,
JQ DnO

which is bounded by / N2^) da.

Jan
I I I = - f Su^DnS——
JH DnO

= - I div [^——WV] - VIV . V ( ^ )

J^ L D^O J \L>nO/

= f [2VA^ . Vu-6 - VIV . (,^)^

2
+ VA^ . W - 2 ] .
Yn I- Dn6 \{Dn6) / Dn6\

The first integral is bounded by (||^V(n)||^2) • (||5'(zt)||^2) and the second is

bounded by ||A^(^)||^2. For the third term, we use the extra fact about Dn6^
namely that VDn6 = {Dn6)2 ^ V^^-, where each {DnSj^S'1 is Carleson.
3
Let us replace \7Dn6 by the expression (D^^V-Dy^ and integrate by parts
once again :

/ VDn6 ' V6u2Dn6 =- ( D^^u^D^ -2 f Dn6^6 • VuuDnS

Jfl Jfl. J^i
- [ Dn6u2V6'\/Dn6+f Dn^S^DnSda.
Jfl Jan
The first integral is dominated by

( f IIVI^-^V /
. f / u^S^Y /
<C ( N2{u)d^
^Jn ^Jn Jan
using both of the Carleson measure properties of 6. The remaining integrals
are handled similarly. See Theorem 2 of §3 below.
We note that the first equality in (1), which relates the square
function to the differential operator and pertains to the coercivity problem
mentioned above, is generalized by line (14) of §3 below. We also note that
the Lp equivalence (p ^ 2) between nontangential maximal functions and
square functions of solutions can be obtained via the powerful technique
of good-A inequalities ([6]). We briefly sketch the arguments in the higher
order situation in §4.
We shall now point out some important applications of Theorems
1 and 2. The first concerns the weak maximum principle in non-smooth
domains. It was shown in [26] that a weak maximum exists for gradients of
biharmonic functions in Lipschitz domains in M2 or R3 or in C1 domains
in R71, any n. That is, if AAn = 0 in ^ and 7v(Vn) € ^(Ar) then in
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1431

fact ||V||^oo(^) < C\ | V^| |L°O(^). (This was later proved in [28] for ^rn~lu
for A^ = 0 in such domains.) As noted in [26] such an estimate has
new consequences for the classical Dirichlet problem posed for continuous
data. The main ingredients necessary to derive weak maximum principle
are the following. First, one requires L2 solvability of the Dirichlet and
Regularity problems (see [27] and [33] for higher order equations and
systems). Second, and most important, one must extend solvability of the
Regularity problem to p near 1. In three dimensions, for second order
elliptic systems, this is first done in [13]. These ideas were used in [26]
for the biharmonic equation. Finally, one needs an LP relationship between
solutions and their Riesz transforms for p near 1. This can be obtained
from the Lp equivalence between nontangential maximal functions and
square functions. The observation that these three ingredients lead to weak
maximum principles was first made in [26], but see also [13], [4] and [29] for
other applications. For a sketch of the proof of the weak maximum principle
for higher order systems see [33].
Precise Sobolev/Besov space estimates on solutions are a second
important application of square function estimates. These in turn lead to
the existence and uniqueness of inhomogeneous Dirichlet problems for these
higher order equations and systems of equations. Such a program has been
carried out in the case of Laplace's equation in Lipschitz and C1 domains in
[19] and for the biharmonic operator in [1]. These inhomogeneous problems
with Sobolev space data reduce to homogeneous problems with data in the
appropriate trace spaces. One of the key ingredients in carrying out this
program is determining precisely which Sobolev spaces tie the solution to
Dirichlet or regularity problems with LP boundary data. Thus, for example,
the estimate
f 6(X)\\/mu(X)\2 dX <-^oo
Jfl.
for solutions u to 2m-order elliptic operators, together with interior esti-
mates implies that V771"1^ belongs to the Besov space A^Q).

1. Preliminaries.

We define an elliptic symmetric J^-system as follows. By D we

denote the vector of first partial derivatives ( — — , . . . , ——) with X =
\9X\ 9Xn!
( X i , . . . , Xn) e R71. Let L^ = ^ D^a^D^, for m, fc, and I positive
H=|/3|=m
1432 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

integers, a = (ai,... ,0n) and /? = (/3i,... ,/?„) multi-indices, and a^

real constants. Let 1/^(0 = ^; ^a^^ for $ € M71. Then L is a
|a|=|/?|==m
K
symmetric JC-system if Lu = ^ L^V for -a = (u1,' • • , ^), k = 1, • • • K
1=1
and L^ = L^. If in addition, we assume that the Legendre-Hadamard
condition holds : Re ^ ^(00^ ^ ^I^ICI 2 for some E > 0, C =
fc,Z==l
( C i ^ - ^ C n ) ^ C", $ e y, then L is a (strongly) elliptic symmetric K-
system. When K = 1, L is in general a single higher order elliptic equation.
^2 C R71 is Lipschitz if there are a finite number of neighborhoods
that cover the boundary 9^1 so that within each neighborhood, 9fl may be
given as the graph of a Lipschitz function (p : R71-1 —^ R. The maximum of
the Lipschitz norms of each (p together with the number of neighborhoods
essentially describe the Lipschitz nature of fL See, for example, [9], [27].
Dirichlet data will be taken in ^(9^2) spaces with respect to surface
measure da where in general 1 < p <, oo. Thus boundary values will be
taken in the sense of nontangential convergence a.e. da. To do this we define
nontangential approach regions for each Q € 9D
r(0) = i\(Q) = [x e ^: \x - Q\ < (i + a) dist(x, 9^)}
where a > 0 is taken large enough depending on the Lipschitz constant
associated with ^l. Then if X—>Q
lim u(X) = g(Q) we say that u has
xer(Q)
nontangential limit g{Q) at Q.
The nontangential maximal function of a function u in Q is defined by
N{u)(Q) = sup \u(X)\.
^er(Q)
Let f2 be the domain above the graph of a Lipschitz function y? :
R71"1 —^ ]R. We will need to recall the adapted distance function associ-
ated to ^ introduced in [11] as well as a version of this defined by [21].
i ij>
If F(x,t} = ct + rft * (p(x), where rjt{x) = —r^-) is a smooth com-
~c i,
pactly supported bump function, and if c is sufficiently large (depending
9F
only on ||V(/?||oo), then — > c' > 0 for c' = c'(||V^||oo). Thus, every
X e O can be represented as {x,F(x,t}) for x G R71"1, t > 0, and one
defines 6(x,F{x,t)) = t. Then 6(X) w dist(X,<^2), Dn6(X) > C" > 0,
|V5(X)| < C 1 " , and 6(X)\^6(X)\2dX is a Carleson measure. It is easy
to see that t|WF(.r, t^dxdt is a Carleson measure and the Carleson prop-
erty of 6 follows from the relationships between the derivatives of 6 and of F.
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1433

Transverse differentiation of F or 6 has an additional

r\ important property.
If we define ^(y) = ^Q/), then D^y) = - ^ —-^Q/), so that if T] is
i ^Vi
chosen appropriately to be radial and even, f ^ ( y ) = 0 for each i and so
-|^ * b(x)\2dxdt is also a Carleson measure for any b € L°°. In terms of 6
~c
this means that ^/Dn6 may be replaced by terms of the form \7Dn6 where
6 has the properties that \DnS(X)\26~l{X)dX is a Carleson measure and
\Dn^\ is uniformly bounded depending only on ||V^||oo. In fact we may
r\

take Dn6(X) = D^-(X) = ^ * -^-^(x), (1 ^ j < n-1). These facts will

°yj
prove useful in proving Theorem 2 of §3. In general the Carleson measure
properties of 6 that will be used take the form

/ \u\p\^V6\<26dX <C ( N(u)Pda

J^i JQ^I
and
( ^{DnSfS^dX <C { N(u)Pda, 0 < p < oo.
Jo, Jan
The square function for a function u in f^ C M71 is defined by

l/
s^m={l
\Jr{Q} ^&«'^
i^ - yi / '.
For solutions to elliptic symmetric J^-systems of order 2m Lu = 0 it
will be established in §4 that for any 0 < p < oo

{ N^^uYda
JQ^I
is equivalent to
( S^^^Pda.
JQ^I

When p = 2, by Fubinfs theorem, this equivalence is the same as the

equivalence between
{ ^(V771-1^)2^
JQ^I
and
{ IV^X^disHJQdX.
JQ
JQ

This latter equivalence is proved in §§2 and 3.

1434 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

The following lemma is a standard interior estimate for solutions v of

-ftr-systems Lv = f ([15], p. 517, [3]) using L2 averages. To express it in this
way requires an interior estimate assumption on / itself, which hypothesis
will be met in the cases of interest to us. The distance of X to 90, will be
denoted dist(X).

LEMMA I (Interior estimates). — Let L be a 2m-order K-system as

above and let v 6 C^2771^) satisfy Lv = f on f2 C R71, a Lipschitz domain.
Suppose further that supjV^/l < Cr~k^:B2r\9\ where g satisfies an interior
Br
estimate : 3 C\ > 0, so that

sup|^|<Gi(^|^| 2 )^ forB.cn

with dist(-B3y.,9f2) w r. Then, for some 03 > 0 independent of v, but

depending on the Lipschitz character off2,

/ dist^+^IV^^X)!2^
Jo.
<c^1 I |Vm^(X)|2dist(X)dX+ [ dist^^X^gW^dx}.
)
JQ J^i

LEMMA II. — Let fl. C S^ be a bounded Lipschitz domain and fix

a point P* € ^. Let F 6 C1^) with F(P*) = 0. Then for any £ > 0 there
exists Ce < oo depending only £, P* and the Lipschitz nature offl, so that

[ \F(X)\2 dist(X)dX <e [ N^do- + Ce [ IVFpdis^JQdX.

J^i JQQ Jfl

Proof. — By [24] (see for example [31]) ^ may be approximated by

domains f^' C ^2 with Lipschitz nature the same as that of ^2 so that
there is a homeomorphism A : 90. —> 9^1' with maxgean 10 — A(Q)| as
small as we want. Choosing an appropriate 0.' the first term on the right
dominates f^\^, \F\2 dist(X)dX and by the standard Poincare inequality
the remainder of the left side may be dominated by the second term on the
right. D

LEMMA III. — Let fl. C M71 be a bounded Lipschitz domain and let
F G C^^). Then there exists a Lipschitz domain f2 C f2 and a constant
C with C and dist^',^) > 0

(1) f N^da ^C { N^F^da + C ( F^dX.

JQ^. JQ^. Jw
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1435

Proof. — By the fundamental theorem of calculus we have for each

Q € 9^
N(F){Q) < \F(Q)\ +C77V(VF)(Q).

Similarly with a domain n' and homeomorphism A as in the proof of the

last lemma we get
|F(0)|<|F(A(Q))|+C7V(VF)(Q).
Thus,
/ N(FYda<C f N^F^da+C { F2daf.
JQ^I Ja^i JQQ,'
Now one can average with respect to a continuum of such domains ^ and
obtain (1). n

2. Square function dominates maximal function in L2.

The essence of the argument showing that the L2 norm of the

nontangential maximal function is dominated by the L2 norm of the square
function is in the proofs of the next two lemmas.

MAIN LEMMA 1. — Let fl, be the domain above a Lipschitz function

(R : yi-i __, ^ ^ ^ ^ ^ C2771^) be compactly supported in R71 and
satisfy Lv = f for any elliptic K-system L of order 2m as in §1. Let
6{X) be the adapted distance function of§l. Then there is a constant C
depending only on ||Vy?||oo5 m, n, K and E so that

f ^^v^Nnda < C\L ( IV^I^JQdX

JQQ JQ
+ / ^^^WdX
J^i
(1) + ll^(Vm-l^)||^(^){(y> IV^dist^x)'

+vf / |^|2^m+l(^x) 2
/ ) }+ / l/IIV^^I^Wdxl
^ J^. -I
where g is as in Lemma I of §1.

To prove this we will repeatedly apply the inequality established

below.
1436 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

LEMMA 2. — With the hypotheses of Lemma 1 and multi-indices

\a\ = m — 1 and \{3\ > m,
| / D^v Dftv{Dr,6)rn-^6^-mdX
U^
<c\L [ |Vm^|2dist(X)dX+ f \g\262m^ldX
^ J^

(2) + HA^V—^)!^) . ( / > ^(X^distWdx)12

+ HA^V—^)!^) • ( / ^s^dx)12
+ I / D^^D^^^)771-1-^!^^^1-7"^!]
'JQ U
where ^ is the upper bound for Lv = f satisfying the assumptions of
Lemma I of§l.

Proof. — We write the integral on the left side of (2) as

(1 + \f5\ - m)-1 f Da^D^(^^)m-l-^l^^l+l^l-m)dX
J^.
= - (1 + |/3| - m)-1 / D'^DnvD^vWr^-^S^^dX
Jn
+ / DavDftv{Dn6)rn-2-WD2,661^-rndX
Jfl

- (1 + \(3\ - m)~1 { D^vD^DnV^D^-^^S^^dX.

Jo.
By the Schwarz inequality the first integral is bounded by

(3) ( f IV^I^dx) 2 ( f IV'^I2^!-27^1^ 2

\Jfl ) \Jn )
By the lemma on interior estimates (3) is bounded by
(4) C f ^v^SdX + C I |p|2^m+ldX.
JQ J^i
By the Schwarz inequality, interior estimates and the Carleson mea-
sure property of 6 the second integral is bounded by IIA^V771"1^)!!^^)
times the square root of (4). n

Proof of Main Lemma. — By the Gauss divergence theorem the left

side of (1) is equal to a sum of terms of the form
/ D^vD^DnV dX where H = m - 1.
Jfl.
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1437

We apply Lemma 2 a total of m times in order to repeatedly bound

the last term of Lemma 2. The result is (1) with

(5) / D^vD^D^lv(Dn6)-rn6m dX
JQ
in place of the last term in (1).
We next remark that on the boundary of a domain one may inter-
change the indices of components of the unit normal vector with the indices
on spacial derivatives by introducing tangential derivatives on the bound-
ary as the following typical calculation shows:
N ' N ^ D j D ^ v = N^N^DiD^v + A^(A^- - NW^D^v
(see the remark before Theorem 3.3 in [27]).
Hence define the symmetric K by K matrix A = A(X) on all of fl
to have entries Y, Na(Xya^N^(X) 1 < k,l < K where we write
|a|=|/3|=m
L^ = Y, D^a^D^, the L^ as in §1. Here N(X) is the normal at X ,
|oi|==|/3|=m

viz. on the level surface {6 = S{X)}. By the Legendre-Hadamard

IVo^AJl
condition A has a uniformly bounded inverse for all X and we rewrite (5)
as

(6) / A-\D^v) • A(D^D^lv)(Dn6)-m6rndX.

Jfl.
Next for each 1 < k < K the components of the vector obtained from
operating with the matrix A may be analyzed by
K
(7) E E N^yD^D^v1
l=\ |a|=|/3|==m

=E E ^a^D^AP(A^)m+l+
^=1 |a|=|/3|=m

terms with one tangential derivative on 2m — 1 spacial derivatives of v.

This follows by 2m applications of the "typical calculation" mentioned
above. The last term in (1) is now obtained from the first term on the right
of (7) when inserted in (6).
The tangential derivative terms of (7) all typically look like the last
term of the "typical calculation" with \v\ = |/^| = 2m — 1. Since the
tangential derivative is along the level sets of 6, transferring it by integration
1438 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

by parts to the other functions in (6) results in either one more derivative
on D^v or a second derivative on 6 when an N or Dn6 or coefficient from
A~ 1 is differentiated. Since Dn6 is uniformly bounded away from zero and
|V<5| is uniformly bounded from above, (6) now yields two more types of
integrals which may be bounded by

f iv^l |V2m-l^mdX+ / ^rn-lv\ IV^-^I IVV^dX.

^ J^
An application of Young's inequality and the interior estimate of Lemma I
with i = m - 1 to the first integral yields the first two terms on the right
of (1). By the Schwarz inequality the second integral is bounded by

( ( IV 771 - 1 ^! 2 IVV^X)^ f IV27"-1^2771-1^)'

which by the Carleson measure property of 6 and Lemma I again yields the
third summand on the right of (1). Q

We may now prove the theorem of this section

THEOREM 1. — Let 0 C R71 be a bounded Lipschitz domain with

connected boundary. Fix a point P* e ^2. Let L be any elliptic symmetric
K-system of order 2m from §1. Let u be any solution to Lu = 0 so that
^(P^V^P*),...^771-1^?*) vanish. Then there exists a constant C
depending only on the Lipschitz character off2, n, m, K and E so that

(8) f N^-^u^da ^ C { IV^X^dist^dX.

JQ^I J^

Proof. — Since it suffices to establish (8) on domains approximating

^ from the inside as in Lemma II with a constant independent of the
approximation we may assume that all nontangential limits for u and its
derivatives exists on 9fl. Consequently by [27] (see [33] for systems) the
left side of (8) is dominated by C f^ {^^u^do- with C depending only
on the Lipschitz nature of f2.
By definition of bounded Lipschitz domain, 90, may be covered by a
finite collection of open right circular truncated cylinders {Zj} where each
Zj is centered at a boundary point, has a lateral diameter diamZ^, the
concentric double 2Z^ is associated with a rectangular coordinate system
and Lipschitz function ^ : W1-1 -^ R such that 2Zj H ^ = {(x,s) :
^pj(x) < s < 20||V^||oodiamZ^, |.r| < diamZj, and 2Zj n ^ = {(x,s) :
s < (pj(x)}r\2Zj.
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1439

For each cylinder let \j be a smooth cut-off function supported in

2Zj, identically equal to 1 in Zj, and so that IV^^X)] < C^diamZ^)-^
for all k > 0 and X € M72. Putting Uj = \ju we will prove for any e > 0
and each j

(9) [ IV771-1^-!2^ <C, ( ^u^X)}2 dist(X)dX

JQ^r\2Zj Jfl

+e [ N^-^u^da
Ja^i
where Ce < oo depends only on e and the Lipschitz nature of f^. By
choosing e small enough depending only on the Lipschitz nature of f^
(8) follows. We have Q^l H 2Zj c Kj = {(x^j(x)) : x C W1-1} with
||V^j||oo < M where M depends only on the Lipschitz nature of ^ and
suppn^ c {(x,s) : s > (pj(x)}. Let N = ( N 1 , . . . ^ n ) denote the mner
unit normal vector on Aj.
Then the left side of (9) may be dominated by

/ ^-^da^C ( IV7"-1^!2^7^
J9^tn2Zj JAj
where C depends only on M. Applying the main Lemma, with / = Luj
and I/I < |^-| since Lu = 0 implies we may take
2m
g, = C^diamZ^IV27"-^
k=l
we obtain
/ iv771-1^!2^
J9^n2Zj

<C\ ( |Vm^,|2dist(X)dX+ [ \9j\2 dist^^WdX

'-Jn Jo.
(10) + ||^(Vm-l^)||^^)((/> IV^dist^x)'

+(^|^|2dist(X)2m+ldx)')

+ /> |L^| IV^-^ldist^JOdxI.

| -L/ ULj | [ V L(-j j diSt (

*/^ /^ -I
That gj satisfies the interior estimates in the hypothesis of Lemma I is
standard (see e.g. Lemma 4.1 of [27]). The first term on the right of (10)
contains terms dominated by
2
k..\2^
f |V^ dist(X)dX k = 0,..., m.
Jfl
1440 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

By repeated application of Lemma II these may all be dominated by

m-l . „

(11) ej" / N{^ u) da+C, \ IV^I 2 dist{X)dX

k 2

k=o JQ^ J
^
for any e > 0 and Cg as in the statement of Lemma II. This in turn by
repeated use of Lemma III may be dominated by
/. m-l .
£ N(^rn~lu)2da + e V" / ^u^dX
JQ^I ^ J^i"

+Ce 1 IV^I 2 dist(X)dX

J^
for any e > 0 where ^2" C ^ is as in Lemma III. Now the vanishing at P*
justifies the Poincare inequality and we get for any e > 0

C { IV^Idis^JOdX <,e ( N^^u^da

Jfl. JQ^I
+C, I ^u^distWdX.
J^
Standard interior estimates [20] may be applied to the first m — 1
terms in the definition of gj to reduce the second term on the right of (10)
to the first.
The IIA^V77'1-1^)!!^^) factor in the third term on the right of (10)
may after repeated application of Lemma III be dominated by
m—l „ j_
CII^V—^H^^+G^ ( / IVSl2^)2
JQ
A;=0

and then via the Poincare inequality by just IIA^V771'"1'^)!!^^)- The

second factor is dominated by the square root of the right side of (11).
Applying Young's inequality to the last term on the right of (10), it
may be dominated by, for example,

C { l^fdistpQ^-^X+G / IV^^fdistpOdX.
Jfl. J^t.
Now there are just enough powers on the distance function so that the gj
integral can still be dealt with as described above and the other term is
lower order.
All together an inequality of the type (9) follows. D
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1441

3. Maximal function dominates square function in L2 - Part 1.

Let L = (^)^=i be an elliptic symmetric J^-system homogeneous

of order 2m. When K > 2 one may define adj (L) to be the J^-system
of cofactors of (L^). Then adj(L) is symmetric and homogeneous of order
2(JC—l)m. The Legendre=Hadamard condition for adj(L) follows from that
for L and elementary properties of positive definite symmetric matrices.
Likewise define det(L) to be the single elliptic operator homogeneous of
order 2Km obtained by taking the determinant of (L^). Then
Ladj(L) = adj(L)L = det(L)J
where I is the K x K identity operator.
If Lu = 0 in fl, we want to show that

(1) { IV^I 2 dist(X)dX < C [ N(yrrl~lu)2do•

J^i JQQ.
with C depending only on the Lipschitz nature of fL We will show in this
section that (1) follows for 7^-systems, K >_ 2, under the assumption that
it holds, for scalar solutions to single homogeneous equations of order 2m
for any m. In particular we will assume that (1) holds for the operators
det(L) with m replaced by Km.
Denote the fundamental solution matrix for L by F(X) [20], pp. 75-76.
Put At = adj(L) and M = det(L)J and denote the fundamental solution
matrices for J\f and M by FV and T^ respectively. Then LTj^ = FJ\f.
Given a solution Lu = 0 in fl, and a multi-index |a| = m define

(2) w(X) = f T^(X - Y)Dau(Y)dY.

JQ
Then
Afw = D^u

and
Mw =0

in 0
Since the components of w and all derivatives of those components
each satisfy a single elliptic homogeneous equation of order 2Km^ our
assumption yields

( IV^-^wpdistWdX^ C { ^(V2^-1^-^)2^.
J^i JQ^I
1442 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

Since J\f is of order 2(K — l)m,

( ^u^2 dist(X)dX = [ |A/w|2 dist{X)dX

JQ Jo.
<C ( N^^-^-^da
JQQ.
and (1) follows from the following lemma by letting W = D^w for any
|^| = 2(K - l)m - 1 and w defined by (2).

LEMMA. — Let L be an elliptic symmetric K-system homogeneous

of degree 2m and Lu = 0 m f^. Let B be a K by K matrix of functions
such that V^^-X) = -^2mB(X) and "V^B^tX) = tl~n'V2mB(X) for
any t > 0 and all 0 7^ X -^ R71. Let \a\ = m and put

W{X) = ( LB{X - Y)Doiu(Y)dY, X € ^.

JQ
Then there is a C < oo depending only on L, B and the Lipschitz nature
offl, so that
(3) \\N(W)\\^w <. W^-^h^.

Proof. — Let ^ be smooth cut-off function supported in a covering

cylinder 2Z such that \ = 1 in Z. Define u == \u and

(4) W(X) = I LB(X - Y)Dau(Y)dY.

J2Zr\^t
It suffices to obtain (3) with jZ D Qfl, in place of 9^1 on the left side.
Hence for Q e \Z^Q^l, N(W)(Q) < N(W){Q)+N(W-W){Q) and
the last term is easily controlled by
m-l

^Edi^i^^+ii^^n^w)
k=0
since a single application of the Gauss divergence theorem yields
\(W - W){X)\ ^ C { ——————\^-^^ _ n)(p)\dp
Ja^i\z \^ ~ ^r

+c Jo.\z
I \v|A -^ini^"
ri'
1
1^ - ^w\dy
and (1 — \)u is supported uniformly away from X when X is in the
truncated cones at Q.
To control N(W) define, as on p. 16 of [27], the mth primitive u-rn
of u so that D^u-m = u and u-rn is also supported in 2Z D fl. Here a
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1443

rectangular coordinate system has been chosen so that the derivative Dn

is transverse to 9^1 H 2Z. Then integrating (4) by parts to transfer the
operator L onto u-m and the derivatives DaD^l yields as in line (3) p. 15
of [27]
(5) N(W){Q) < ^(^(V2771-^-^))^)

+ sup | / D^^x-y^-^y^y
xer(Q)' Jiz
where /C represents potential operators on 9^1 that by [8] have the property

11^(^(/)) HLP(^) < Cp\\fh?^ 1 < P < oo.

But now the right side of (5) may be dealt with as in the proof of
Theorem 4.6 of [27] by using Theorem 3.2 and Lemma 4.5 of that paper to
control the first term and Lemma 4.4 [27] to control the second. (See [33]
for a discussion of the extension of these results to symmetric JC-systems.)
Altogether and summing over all covering cylinders
m—l

\\N{W)h^ < C ^ (||V^||^(^) + IIV^H^)).

k=0
Since the definition of W is unchanged upon subtracting any polynomial
of degree m — l from IA, Poincare inequalities on ^l and 9^1 may be justified
to yield (3). D

We remark that it is the proof of this lemma with its reliance on the
Riesz transform results of [27] and [33] where symmetry of our real systems
plays a role. For some of the difficulties encountered with nonsymmetric or
complex coefficient systems see [34] and [35].

3. Maximal function dominates square function in L2 - Part 2.

Part 1 above showed that the inequality (1) holds for solutions to
-ftT-systems if it is known that (1) holds for solutions to single homogeneous
equations of order 2m ,m € Z+. Now we will show that (1) must hold
for these latter solutions if it is known that (1) holds for solutions to
homogeneous equations A4u = 0 where M. is elliptic and of the specialized
form
(6) M = ^ U^)2, a^ > 0 for all |a| = p.
\a\=p
1444 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

The pointwise quadratic form over the Sobolev space H13^) associated
with M. is then
^ aa^u^X))2, da > 0 for all |a| = p
\a\=p

which is clearly equivalent to [V^z^X)!2 so that trivially

(7) f|V^(X)|2 dist(X)dX < C { V" a^D^X^distpOdX.
7
^\o^P

In part 3 below it will be shown how the latter quantity can be dominated
by the right side of (1) plus a term that can be hidden on the left.
As an example of the type of elliptic operators we need to analyze
here, consider in M4 the 4th order operator
L = Di + (DM2 4- (D^)2 + (DM2 - 41)1^2^3^4.
Given any e > 0 Le = L + s(D^ + D^ + D^) is an elliptic operator. L is
derived from a corresponding semipositive definite polynomial known as a
Motzkin polynomial [23]. It has been shown (see also [7]) that L does not
admit a form of the type required by the first hypothesis of the Aronszajn-
Smith coerciveness Theorem [2], p. 161, i.e. that the form be a sum of
squares
f ^^(D^dX
J^:
^j=l
where the Pj(D) are homogeneous constant coefficient polynomials. It is
easy to show that the same is true for Le when e is small.
The Aronszajn-Smith result is apparently still the best available for
obtaining inequalities somewhat more general than (7) with C depending
only on the Lipschitz nature of n (see comments [2], p. 167), but clearly
cannot be used for operators such as the L^.
Instead we rely on a theorem due to Habicht [17], pp. 300-302 on
strictly positive homogeneous polynomials with real coefficients, i.e. any
of our elliptic operators L. By examination of the formula (11.3.2) of [17],
p. 302 one can assert that any elliptic homogeneous operator L with real
coefficients will satisfy
ML = LAf = M
where M. is as in (6) and At is another elliptic homogeneous operator like L.
But now L, M. and M have precisely the same formal relationship as
they did in Part 1 above and one can argue in the same way. We conclude
that
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1445

(1) holds for solutions to Lu = 0 if it is known to hold for solutions to

Mu = 0 with M as in (6).

3. Maximal function dominates square functions in L2 - Part 3.

Consider now elliptic operators L = ^ a^-D20 where a^ > 0 for

|o;|=m
all | a = m. We will establish the inequality

(8) ^ H = ma^ [ (D^X))2 dist(X) dX < C { N^^u)2 da

^ JQ^
for solutions to Lu = 0. As explained in Part 2 above this suffices to
establish (1) for all elliptic K -systems.
As with the model case of harmonic functions, after replacing the
distance function with the adapted distance function 6(X), our immediate
goal is to use integration by parts twice in order to rewrite the left side
of (8) as boundary integrals plus solid integrals with the latter integrals
having two derivatives on 6. In order to do this we will use the following
combinatorial lemma and a modification of it, Lemma 4, below. Denote
the permutation group on m elements by Prn and members of Prn by
(z) = ( % ! , . . . , im).

LEMMA 1. — Let each ofD^,..., Dm denote a differential monomial

. . . , or ——. Let u € C2171. Then
9 X ^ ' " ' ^ 9X

(Dl
••• Dmn)2= 2(^ E ^(A....A^)
2
U)
v /
(z)€Pm

-. m—2

(9) - ., ^
^V
(-l^m-l-fc)
-/ V- - '.,
fc=0

x ^ Di.Di,
(t)ep^
(15,3 • • . D^uD^ ... D^D^... D^u)
+{-l) uD2,...D^u.
m

Here when m = 1 this can be read as

(£»iu)2 = ^u2 - uD\u.

1446 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

The proof of Lemma 1 is by induction on m and uses the next lemma

to take care of an intermediate term.

LEMMA 2. — With the hypotheses of Lemma 1

(10) m ^ D^uD^...D^u
WePm
m—2

=(-i)"1]^ (-1)^+1)
k=0

x ^ DiMD^...D^_,uD^...D^_^...D^u)
(i)6Pm

+ (-I)"1-1 ^ (A,... 1^ u)2 - {m - 1) ^ ^ ... <n.

(l)C'Pm (l)e'Pm

Proof. — When m = 2 this reads

^D^uD^u^D^uD^u) = 2D^D'2{uD^D'2u)-2(D^D-2u)2-2uD^Dtju

which is true.

For the inductive step assume (10) holds for m. Let (j) € Pm+i-
When summed in (j) the identity

D^uD^D], ... D^u + D^uD],D^D], ... D^u

= D^D^uD^D^... D]^u) - D^uD^D], ... D^u
-uD2Jl ...D2Jm+1 u

becomes

(11) 2 ^ D^uD^...D]^u
(j)ep^+i

= E [D^{uD^D^...D]^u)
a)e-p^+i
- D^uD^D], ... D^u] - (m + l)\uD^... D^u.

Next let Pf^ denote permutations of the m elements { l , 2 , . . . , j — 1.

j + l , . . . , m + l } . Then
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1447

m ^ D^D^uD^D^.^D^u
U)^m+l

m+1
D
-^E E ^lD^^D^...D^D,u
^=1 W^m

m+1 m—2

=E[(- ) E(- l ) fc ( fc + l ) E
•3=1
l m
^^ (»)6'P^

A. A. ?3 • • • A^^A.... A^<_^ ... <^)

+(-l)"1-1 ^ (A,...A„0,^t) 2 -(m-l)
(t)e^
^ DjuD2 ...£»? £),d
/ ^/ J rl I'm J •I

^)eP^

where the inductive hypothesis has been used. Thus

(12) -m ^ ^D,^^D,,^...D^n
(j)GPm+l

m—2

=(-ir^^-if(k+i)x ^ ^,^
^=0 0')ePm+i
(D^ ... D^_^uD^ ... D^_^D]_^ . . . D^u)
+ (-l) m (m + l)!(Di... Dm+iu)2 + (m - 1)
E ^^i^L--^^-
(j)ePm+i

Multiplying (11) by m, substituting (12) for the second term under

he summation on the right side, combining like terms, and noting that the
first term on the right side of (11) becomes the k = m - 1 term of (10) we
obtain (10) with m + 1 in place of m. D

Proof of Lemma 1. — The m = 2 case reads

(D^uf = ^(D^u)2 + DJ(Dm) 2 ) - DMuD^u) + uD^u.

1448 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

Assume (9) holds for m. Then

^(m+l)!^...^^)^^ ^ (^...^^)2
(.7)<EPm+l

-. m+1

=!n^E E^...^,.)2
J
= l (^)e^

, -. m+1

= -„,- E [7 E D?! (D^ • • • A^u)2

7 7 1 + 1 \-^ [171 \—s 9 . ,<->

-'=1 (*)€-P^
TO-2

-^(-^(m-l-fc) ^ DiMDi,...D^_,D,uDi,...D^_,
k=o
WeP^
2 1
D-"m.—fc-|-l
. . . . £»?
'"m
£>,u)
J '
+ v(-1)"
/
Y
^_^
D,uD?
J ii\
...£»?"m. £»-J
•/ |

(z)eP^
=m
^ E ^(^•••^.^)2
0)€Pm+i
m
1 —2

-^E^^-i-^) fc=0

D
x E ^D^D^•••D^-^D^...D^_^D]_^...D^u)
0)€Pm+l

^(_^-^ ^ ^,^,^...D^n
a)epm+i

where the third equality uses the inductive hypothesis. Dividing by (m+1)!
and using Lemma 2 on the last term we obtain

"^(^...D^)'^ S 0?,W,...^,«)2
v /
U^m+1

. -i .. m—2
m+1 1 ^-^ r.
-———7———r-> (-^"(m- 1 -k)
m (m+1)! 2 ^ v / v /

D
x E nDn^.^D^_^uD^...D^_^D]_^^^
(j)CPm+l

--7—^Ty(- l )^+ l )
1

m(m+1)!
1 m-l

2:
-^v / v /
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1449

x E D^D^D^-'D^-^D^...D^_^D^_^...D2^^^^^
(j')ePm+i
+ ^(Di... D^u)2 + {-l^uDi... D^u.

The two summations in k combine to yield the summation in k of (9)

when m is replaced with m + 1 and the lemma follows. D

The equation Lu = 0 will be used to eliminate the last terms arising

from the application of Lemma 1 to the left side of (8). The remaining terms
allow us to transfer two derivatives to the (adapted) distance function.
Unlike the first term, however, the middle summation does not directly
yield boundary terms like the right side of (8). To remedy this we resort to
primitives as in the proof of the lemma from part 1 above. Here v-k will
denote the fcth primitive of v and 9 will represent the derivative such that
Qkv-k' First a preliminary lemma.

LEMMA 3. — With the hypothesis of Lemma 1 and v € C2771, m > 2,

(13)
m—l

^(-l)'-1 ^ 9D^D^...D^_^uD^...D^_^
k=l (»)€-Pm

^-^•••<^-l)
m

=EK-l)m-^2 E 9k-2uD^...D^v.k
k=2 (i)e-Pn,

+9 2 E ^...D^_,9k-2uD^...D^_,D^...D^v.k
WePm

- E 9D^Di,...D^_,9k-luD^...D^_,
(06'Pm

Df^.-.D^)}.

Proof. — The m = 2 case reads

QD^uD^v^) +<9Di(nDJDi^-i) = 292(uD<iD^v^)
+ Q^^D^uD^D^^ + D^uD^D-iV^)
- QD^QuD^D^^) - QD^QuD^D^v^)'

We will establish (13) with m+1 in place of m assumming (13) holds

1450 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

as written. Hence
771
1 1
E^
^=1
)'" (z)ePm+i
E
9A. (^ ... D^_^uD^... A_^<_^ ... <^-i)
-(-I)7-1 E ^(^t-^L^)
(^)ePm+i
m-1
fc l
+E(-l)
fc=i
- WePm+i
E
9\D^ ... D^_^uD^D^... A..^<_^ • • • <,^-2)
+ ^ a2^ ... D^uD^... A^^-2)
(z)CPm+l

771
l fc l
+E(-
A;=2
) - {i)ePm+i
E
9\D^ ... D^^uD2^... D^_^D^... <^-2)
- ^ 9D^D^...D^9uD^D^...D^v^)
^)ePm+i
7n

+E(-1)' E
fc=2 (»)6'Pm+l

9D^(A3... D^_^9uD^... D^_^D^... <^_2).

Here we have used the identity

9D(UV) = 9\UDV-^ + 9\DUV-^ - 9D{9UV,^

to each term on the left, obtaining three sums on the right from which we
have isolated the mth term, the 1st term and the 1st term respectively.

Using the fact that we are summing over the permutation group, the
first and second summations in k now add to zero.
771+1

Replacing ^ with ^ ^ as in the proof of Lemma 2 and k with

^n+l 1=1 P^

k + 1, the third summation in k may be calculated using the inductive

hypothesis with 9u in place of u and Dfv^ in place of v^. The third
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1451

summation becomes
m

^[(-l)-^2 ^ ^-1.^... <^-,

^=2 (z)e'Pm+i
2
+9 ^ A.... A^_^-1^?^ ... A^<^
(^ePm+i

•••^1^-1
- ^ 9D^D,^^D^_,9kuD^...D^^D^
(^)ePm+i

•••<..^-i)]-

Now replace k with A; - 1 and the three isolated terms supply the
missing k = 2 terms. Q

The next lemma gives us what we need when u = v.

LEMMA 4. — With the hypotheses of Lemma 3,

m-2
^ ( - l ^ m - l - f c ) ^ D^D^D^...D^_,uD^...D^D^_^
k
=o (t)6'Pm

...£»?^TU u)'
m
l fc 2 2
=^(fc-l)[(-l)" - ^ ^ ^- ^...^^
^ (i)eT'^
2
+9 ^ A,...A^_^fc-2^A....A^-^L^...<^
(i)e'Pm
- ^ 9D^D^...D^_,Qk-luD^...D^_,D^
(»)ePm

•••^^)]

m-1
+ ^ (m - k) ^ QD^D^... D^^Q^uDi,... D^_^
k=l (i)e'Pm

<-^--^L^)
m-1

- ^ ( m - f e ) ^ D,MD^...D^_^9kuD^...D^_^D^ ^
fc=i (i)ePm

•••-DL^-fc).
1452 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

Proof. — The m = 2 case reads

2D^D^uD^v) =292(uD^D^v^)
+ 9<2(D^uD^D2V^ + D^uD^D^v^)
- QD^QuD^v^) - OD^9uD^v^)
+ QD^D^uD^v^) + OD^uD^v^)
-2D^(9uD^ D^v-i).

Using the identity

DMUV) = QD^UD^ + D^UV-^) - DMQUV^)
and summation over the permutation group, the m+1 case can be written
m—l

^ (-l^m -fe) ^ ^i^(A3...A^_^A....A^_.

^=0 (»)ePn.+i
<^---<.^)
m

=E(- l ) fc-l E 9D
^Di.•••D^-.uD^...D^_,
k=l (»)6-Pm+l

^^-.•••^x^-l)

+m ^ ^A,(^A3...A^uAi...A^^-i)
(»)€•?„,+!

-m ^ A,A,(£'z3--.A^9uA,...A^^-i)
(^e^^+i
m-2

+^(-l)k(m-l-k) ^ D^D^D^...D^9uDi,...D^_,
k=0 (t)€Pm+l

<^--.<^-l).

Now Lemma 3 is used on the first term and the inductive hypothesis
as in the proof of Lemma 3 on the last term. D

Combining Lemmas 1 and 4 we obtain

(14) (D^)2 = - mQ^uD^u,^ - (m - l^uD^u.m
+ terms of the form
^2^m-lu\7m-l+ku^) for k = 0,..., m
whenever |o;| = m. With this observation we can prove the main lemma
of this section. Let Q be a bounded Lipschitz domain and let Z denote
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1453

a covering cylinder with associated cut-off function \ as in the proof of

Theorem 1.

LEMMA 5. — Let L = ^ daD^ as described above and Lu = 0

[c>!|=m
in a neighborhood of 2Z D Q. Let ^ be a cut-off function like \ except
supported in (1/4)Z and identically 1 in (1/8)Z. Then
(15) V da [ (Dau{X))2^{X)dlst(X)dX
l/
W=m "

<, Ce t N(yrn-lu)2d(T+e ( IV^X)!2 dist(X)dX

Jafln'zz J2Zr\^t
where Cg depends only on the Lipschitz nature offl. and e > 0.

Proof. — As in §2 it suffices to obtain our estimate for approximating

smooth domains fL
Define u = \u and primitives u-k as the proof of the Lemma from
Part 1 where a coordinate system has been chosen so that Dn is transverse
to 9^ D 2Z. Let 6(X) denote the adapted distance function of §1 with
respect to the coordinate system. Then 6 may be substituted for dist on
the left of (15) and the result is by (14) equal to
- m [ D^uLu-rn+i^dX - (m - 1) ( D^uLu-rn^dX
J^i J^i
+ terms of the form / V^V^^V^"1"^-^)^^ k = 0 , . . . , m.
J^i
Here because of the choice of coordinate systems Dn plays the role of
9 from (14) while L is actually a rotation of the original L. That L is no
longer in precisely the form (6) is of no consequence.
The first two terms are lower order by virtue of Lu = 0. They may
be analyzed, for example, by line (7) on p. 18 of [27] together with interior
estimates like Lemma 4.1 of that paper. The result is an upper bound of
m—l ,.

(16) GV / ^u^dX.
J^Q Jzno.
For the main terms, whenever a derivative lands on the cut-off
function '0 when using the divergence theorem the result is a lower order
term no worse than f^ ^^-^u^^u-k^dX.
Such a term is also bounded by (16) using the Schwarz inequality,
Hardy inequalities like Lemma 4.2 of [27] and interior estimates like Lemma
1454 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

4.1 of [27]. Thus in the following we will disregard any terms arising from
derivatives falling on ^.
Hence, applying the divergence theorem twice to the main terms leads
to boundary integrals of the form
I \^m-lu\ IV771-1^-^ \^6\da
JQ^\Z

and solid integrals of the form

(17) ( V^uV^-^-^V^dX.
Jfl.
By the Schwarz inequality and Corollary 4.7 of [27] the boundary
terms are bounded by (16) plus
(18) ^IIV^nl^zn^)-

For the solid integrals we introduce the term —r— into the integrals
Dnb
(17) and apply the divergence theorem removing the Dn derivative from the
numerator of the introduced term. This results in three types of integrals
I = { v^^V^^-^V^——dX,
Jfl DnO

m l m-l+fc 2
1 1 = f ^(V - nV ^-A^V ^——dX, and
Jn ^nO
m l m l+ 2
HI = f v - ^V - ^_^V ^^———,dX.
J^l (-l^nO)

The Schwarz inequality allows one to apply the Carleson measure

property of 8 in III to obtain a bound of the type
II^V—1^-,)!!^^^^, k = 0 , . . . ,m
which in turn is bounded by (16) plus (18).
The product rule followed by applications of Young's inequality,
ab < ea2 + — bounds II by the sum of C^ times the bound for III plus
4:e
integrals of the type
e [ IV^^-fcl^dX, k = 0,.... m
Jfl
for £ > 0 to be chosen later.
By Hardy type inequalities (e.g. [27], p. 17) and interior estimates
(e.g. [27], p. 16 these may in turn be bounded by
771 n

eCV \ ^u^SdX.
• Q J2ZW
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1455

For integrals of type I the divergence theorem is used again to remove

a V from V2!^^. Integrals of type II and III are again obtained together
with the new integral type

( ^^u^-^u-k^D^^dX.
7o AlO
Now the crucial fact from §1 that VDy^ may be replaced by VDn6
where 6 has the property that Dn6\26~ldX is a Carleson measure is used.
Using the divergence theorem to remove the V from ^Dn6 now yields a
boundary integral of the type already bound, a type II integral where Dn6
replaces (V2^, and type III integrals where (D^)(VZV) or (I^)(V2^)
replaces (y<26){D2,6)6.
Summarizing we may write
__ „ m—l »

(19) V aa I {Doiu)2^6dX < CA V / ^u^dX

J v
\a\=m ^ J=0 J2^

+ IIV^^II^zn^))
/ +^ / ^u^SdX
J2zno
where Ce < oo and e > 0 may still be chosen depending only on the
Lipschitz nature of fL
By applying a Poincare lemma like Lemma 4.3 of [27] as on page 20
of that paper, the lower order derivatives on u that appear in 2Z and on
2Z D 2f^ may be eliminated and (15) follows. D

THEOREM 2. — Let Q C M71 be a bounded Lipschitz domain with

connected boundary. Let L be any elliptic symmetric K-system of order
2m from §1. Let Lu = 0 in f^. Then there exists a constant C depending
only on the Lipschitz character of^, n, m, K and E so that

(20) f IV^pOpdistpOdX < C ( N^^ufda.

Jfn JQ^I

Proof. — As explained in Parts 1 and 2 above it suffices to prove the

result for L as in Lemma 5. By interior estimates only the portion of f^ near
the boundary need be considered for the left side of (20). The boundary
can be covered by a finite number of covering cylinders <{ -Zj ^ with the Zj
otherwise having the same properties as in the proof of Theorem 1. Since
the left side of (8) dominates the left side of (20) with a multiplicative
1456 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

constant depending only on ellipticity, a finite number of applications of

Lemma 5 yields

/ IV^I2 dist(X)dX < Ce I ^(Vm-l^)2da+^ / [V^l2 dist(X)dX

Jo. JQ^I J^i
where e > 0 may be chosen depending only on the stated quantities,
completing the proof. D

4. Lp equivalence of maximal function and square function.

From Theorems 1 and 2, which give the equivalence between the I?

norms of the nontangential maximal function and the square function of
solutions to Lu = 0 on every Lipschitz domain, a series of near-standard
arguments give the Lp norm equivalence for all 0 < p < oo. In the two
lemmas which follow, we provide the necessary details to prove the good-A
inequalities from which one obtains the next theorem ([6], [16], [10], 12).

THEOREM 3. — Let fl. be a bounded Lipschitz domain in R71 with

constant M and suppose Lu = 0 in 0, where L is a 2m order elliptic
symmetric K-system. Then, if 0 < p < oo, 3C > 0 depending only on
77i, n, TV, M and p such that
C~1 f N^-^uYdcj < f S^-^uYda <C [ N^-^uYdcr.
JQ^. JQ^ Jao
(For the left-hand inequality it is necessary to assume a normalization:
n(P*) = V^(P*) = ... = V^^P*) for some P* € ^.^

LEMMA 1. — If L is a 2m-order elliptic symmetric K-system and

Lu = 0 in fl, C W1, a Lipschitz domain with Lipschitz constant M, then
^7 > 0 sufficiently small and c > 0 both depending only on M such that
for all A > 0, a{Q C 9fl : ^(V^nKO) > 2A, ^(V^nKO) < 7^} <
c72{Q e 9^ : ^(V771-1^)^) > A}.

Proof. — Let A^(0^,r^) be a Whitney decomposition of

{^(V771-1^) > A}.
That is, {Qj} is a sequence of points in {^'(V771"1^) > A} such that
(a) {^(V771-1^) > A} = UA^, where A^-,^-) = B(Qj,rj) H 90.
W SXB(Qj,rj)(^0 < ^ where C depends only on u and M.
3
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1457

(c) Each Bj == B{Qj^rj) is such that 90, n Bj is the graph of a

Lipschitz function and,
(d) 3ro = yo(^) such that if rj < ro, then there is a point Q^ in
B{Qj, 2rj) H ^ with S^^u)^) < A.
As in [12], a sawtooth region Dj associated to
Ej = A^- n {^(V771-1^) > 2A,7V(V m - l ^) < 7A}
may be constructed so that
(i) U (r(Q)nB(o,cir,)) c D, c U (r(Q) nB(Q,c2r,)) for
QeEj Q^Ej
some fixed ci, 03 independent of j
(ii) 9Dj n9^= Ej
(111) QDj is c starlike Lipschitz domain with Lipschitz constant less
than a fixed multiple of M and
(iv) diami^ w rj.
We fix now a particular Aj with rj < r-o. By interior estimates
and property (d) above it can be shown that, for any r > 0,7 may be
chosen sufficiently small so that ^(V771"1^)^) > A/2 for Q € Ej where
Srr- is the square function defined by integration over the truncated cone
r^-r • (Q) = r(0)nJ3(0, rrj). Suppressing the j-subscripts, we wish to shown
that cr(E) < c72a(A). Summing, then on j gives the Lemma. Then we have

(T{E) <— I I ^^-"IV^FdXdaOg)

A JE Jr^(Q)
where 6{X) w \X — Q\. If the aperture of the cones used to define N is
chosen larger than those used to define 5, this last term is bounded from
above by
— ( f dW^^uW^dXda.
A JQD ^r\(Q)
Here r has been chosen sufficiently small so that rV(0) C D for all Q € 9^,
and the collection forms a family of cones for the square function with
respect to the new domain D and d(X) denotes do(X) = (X, 9D) w \X —
Q\. By Theorem 2, this is in turn dominated by —^ J^p ^(V771"1^)2^ <
7 2 a(A), by (iv) and \Vrn~lu\ < -f\ in D. D

LEMMA 2. — Let L be a 2m-order elliptic symmetric K-system and

suppose Lu = 0 in a Lipschitz domain fl, C W1 with Lipschitz constant
1458 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

M and that n(P*) = V^(P*) = ... = V^^P*) = 0 for some P* e ^2.

Then, for sufficiently small 7 > 0 and e > 0, there is a constant c > 0, c, 7, £:
depending only on Jf, m, n and M, such that
^(V771-1^) > 4A,5(Vm-ln) < 7A} H {^(V771-1^) > 7A}^
< ^^{^(V7^-1^) > A},

f where, for any set G C 9^, the set G^ is denned by

G:={Qe^:s^a-^-Gl<e}).
I Q€A cr(A) J

Proof. — We make the a priori assumption that [[^'(V771"1^) ||LP(dcr) =

J? < +00 for some 0 < p < +00, since our goal is to use the lemma to bound
IIA^V7'1"1^)!!? by II^V7""1^)!^. This fact together with interior estimates
on solutions and the normalization on u (the vanishing of V771"1'^?*))
implies that for any compact subset K of f2,
sup {|J^u| : |/3| <, m} <, C(K, R).
X€K

As in [10], this estimate can be used to handle the 'large5 surface balls in the
Whitney decomposition of {N(\/'m~lu) > A}. Thus, let A be one of these
surface balls from this Whitney decomposition (defined as in Lemma 1)
whose radius is less than r^ Set E == A H {^(V^"^) > 4A, ^(V771-1^) <
7A} U {^(V771-1^) >7A}^.
By choosing 7 sufficiently small we can ensure that A^-^V171"1^^ >
2A when Q e E, where Nrr(v)(Q) = sup h^)!- This is straight
^er(Q)nB(Q,rr)
forward using interior estimates and assuming that the square function is
defined with respect to cones with a larger aperture than those used to
define ^(V771"1^). (See for example [10] and then [25].)
Let D be the sawtooth region associated to E. Then, for e sufficiently
small, and Ajc = {Q G A : X e I\-(Q)}, we have

{ ( d^X^uW^dX

\c.ff w^x^^^^^x
L
J JD o"(Ax) >
=Ce t t ^-"WIV^POI^^Q)
^Ah{S(V" ^ - l M)<7A} JT(Q)
<, Cer^^A).
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1459

Recall that D is a starlike Lipschitz domain; let po be the star-

center. We claim that we may assume the normalizations: u(po) = ... =
^m-i^p^ ^ o, by subtracting off a polynomial from n. This depends on
the estimate \^/'m~lu{po)\ <: A + 07 A < 2A for 7 small enough so that if n*
is the normalized n, then A^V^^*) > 2A when A^V771-1^) > 4A. Thus,
for this normalized it, we may use the L2 result of Theorem 1 to obtain

a(E) ^ ^ f N^^-^da

<-
A I N^-^da
JQD
<- I S^^-^da
A JQD
=-^ [ [ doW^-^X^dX

^ ^ 1 1 d^X^-^WfdX < C72a(A).

Summing over the regions E and 6 completes the proof. D

BIBLIOGRAPHY

[1] V. ADOLFSSON and J. PIPHER, The inhomogeneous Dirichlet problem for A 2 in

Lipschitz domains (preprint).
[2] S. AGMON, Lectures on elliptic boundary value problems, D. Van Nostrand,
Princeton, NJ, 1965.
[3] S. AGMON, A. DOUGLIS and L. NIRENBERG, Estimates near the boundary for
solutions of elliptic partial differential equations satisfying general boundary
conditions. II., Comm. Pure Appl. Math., 17 (1964), 35-92.
[4] R.M. BROWN and Z. SHEN, Boundary value problems in Lipschitz cylinders for
three dimensional parabolic systems, Rivista Mat. Ibero., 8, no 3 (1992), 271-
303.
[5] R.M. BROWN and Z. SHEN, Estimates for the Stokes operator in Lipschitz domains,
Ind. U. Math. J., 44, no 4 (1995), 1183-1206.
[6] D. BURKHOLDER and R. GUNDY, Distribution function inequalities for the area
integral, Studia Math., 44 (1972), 527-544.
[7] M.D. CHOI and T.Y. LAM, An old question of Hilbert, Queen's papers on pure and
applied math 46 (1977), Queen's University Kingston, Ontario, 385-405.
[8] R.R. COIFMAN, A. MCINTOSH and Y. MEYER, L'integrale de Cauchy definit un
operateur borne sur L2 pour les courbes lipschitziennes, Ann. of Math., 116
(1982), 361-387.
[9] B.E.J. DAHLBERG, On the Poisson integral for Lipschitz and C1 domains, Studia
Math., 66 (1979), 13-24.
 1460 B.E.J. DAHLBERG, C.E. KENIG, J. PIPHER & G.C. VERCHOTA

[10] B.E.J. DAHLBERG, Weighted norm inequalities for the Lusin area integral and the
non-tangential maximal functions for functions harmonic in a Lipschitz domain,
Studia Math., 67 (1980), 297-314.
[11] B.E.J. DAHLBERG, Poisson semigroups and singular integrals, Proc. A.M.S., 97,
no 1 (1986), 41-48.
[12 ] B.E.J. DAHLBERG, D.S. JERISON and C.E. KENIG, Area integral estimates for
elliptic differential operators with nonsmooth coefficients, Arkiv. Mat., 22 (1984),
97-108.
[13] B.E.J. DAHLBERG and C.E. KENIG, L^ estimates for the 3-dimensional systems
of elastostatics on Lipschitz domains, Analysis and partial differential equations,
Lecture Notes in Pure and Appl. Math., vol. 122, Dekker, New York, (1990),
621-634.
[14] B.E.J. DAHLBERG, C.E. KENIG and G. VERCHOTA, The Dirichlet problem for the
biharmonic equation in a Lipschitz domain, Ann. Inst. Fourier, Grenoble, 36-3
(1986), 109-135.
[15] A. DOUGLIS and L. NIRENBERG, Interior estimates for elliptic systems of partial
differential equations, Comm. Pure Appl. Math., 8 (1955), 503-538.
[16] C. FEFFERMAN and E. STEIN, H13 spaces of several variables, Acta Math., 129
(1972), 137-193.
[17] G.H. HARDY, J.E. LITTLEWOOD and G. POLYA, Inequalities, 2nd ed, Cambridge
University Press, Cambridge, 1988.
[18] S. HOFMANN and J.L. LEWIS, L2 solvability and representation by caloric layer
potential in time-varying domains, Annals Math., 144 (1996), 349—420.
[19] D.S. JERISON and C.E. KENIG, The inhomogeneous Dirichlet problem in Lipschitz
domains, J. Punct. Anal., 130 (1995), 161-219.
[20] F. JOHN, Plane waves and spherical means, Interscience Publishers, Inc., New York,
1955.
[21] C.E. KENIG and E.M. STEIN, Unpublished, communicated by C. E. Kenig.
[22] C. Li, A. MCINTOSH and S. SEMMES, Convolution singular integrals on Lipschitz
surfaces, Journ. A.M.S., 5 (1992), 455-481.
[23] T.S. MOTZKIN, The arithmetic - geometric inequality, Inequalities (0. Shisha, ed),
Academic Press, New York, 1967, 205-224.
[24] J. NECAS, Sur les domaines du type N , Czechoslovak. Math. J., 12 (1962), 274-287.
[25] J. PIPHER and G. VERCHOTA, Area integral results for the biharmonic operator
in Lipschitz domains, Trans. A. M. S., 327, no 2 (1991), 903-917.
[26] J. PIPHER and G. VERCHOTA, The maximum principle for biharmonic functions
in Lipschitz and C1 domains, Commentarii Math. Helvetici, 68 (1993), 385-414.
[27] J. PIPHER and G. VERCHOTA, Dilation invariant estimates and the boundary
Garding inequality for higher order elliptic operators, Ann. of Math., 142 (1995),
1-38.
[28] J. PIPHER and G. VERCHOTA, Maximum principles for the polyharmonic equation
on Lipschitz domains, Potential Analysis, 4, no 6 (1995), 615-636.
[29] Z. SHEN, Resolvent estimates in L13 for elliptic systems in Lipschitz domains, J.
Punct. Anal., 133, no 1 (1995), 224-251.
[30] E.M. STEIN, Singular Integrals and Differentiability Properties of functions, Prince-
ton University Press, Princeton, N.J., 1970.
 AREA INTEGRAL ESTIMATES FOR HIGHER ORDER ELLIPTIC EQUATIONS 1461

[31] G. VERCHOTA, Layer potentials and regularity for the Dirichlet problem for
Laplace's equation in Lipschitz domains, J. Funct. Anal., 59, no 3 (1984), 572-611.
[32] G. VERCHOTA, The Dirichlet problem for the poly harmonic equation in Lipschitz
domains, Ind. Math. J., 39, no 3 (1990).
[33] G. VERCHOTA, Potential for the Dirichlet problem in Lipschitz domains, Potential
Theory - ICPT94 (Krai et al., eds.), Walter de Gruyter & Co., Berlin (1996),
167-187.
[34] G.C. VERCHOTA and A.L. VOGEL, Nonsymmetric systems on nonsmooth planar
domains, to appear, Trans. A.M.S..
[35] G.C. VERCHOTA and A.L. VOGEL, Nonsymmetric systems and area integral
estimates, in preparation.

Manuscrit recu Ie 3 fevrier 1997,

accepte Ie 25 avril 1997.

B.E.J. DAHLBERG,
Chalmers Institute of Technology
Department of Mathematics
Goteborg (Suede).
C.E. KENIG,
University of Chicago
Department of Mathematics
Chicago, IL 60637 (USA).
[email protected]
J. PIPHER,
Brown University
Department of Mathematics
Providence, RI 02912 (USA).
[email protected]
G.C. VERCHOTA
Syracuse University
Department of Mathematics
Syracuse, NY 13244 (USA).