transactions of the

american mathematical society

Volume 290, Number 2, August 1985

THE EULER EQUATION FOR FUNCTIONALS

WITH LINEAR GROWTH
BY
GABRIELE ANZELLOTTI

Abstract. We give a first variation formula for functionals of the type ¡¡¡f(x, p.),
where/(x, p): ß X R* -> R is of linear growth inp for large ¡/?| and ¡i is a Revalued
measure in Ü. The Euler equation for the minima of various functionals defined on
spaces of BV functions is then studied.

Introduction. Let us consider a functional of the type

r r
(1.1) F(u) = / f(x, Du) + / H(x)u(x) dx,

where ß is an open bounded set in R", u: ß -> R and/(x, p): ß X R" -» R, H(x):
ß -» R are sufficiently nice. If a given function u g C2(ß) is a local minimum for
F(u), then u verifies the Euler equation

(1.2) D,fP:(x,Du(x)) = H(x) inß.

On the other hand, in the case that /, H satisfy the assumptions

\p\^f(x,p) ^ M(l +\p\) for all x, p,

,x -.. / is continuous on ß X R",
/is convex inp for each fixed x G ß,
f/ei"(ß) and it is sufficiently small,

the existence of minimum points with prescribed boundary conditions for suitable
extensions [20, 15, 12, 13, 9] of functionals of the type considered is known [17, 15,
12, 1] in the space BV(ß) = [u g Ll(Q)\Du is a measure of bounded total varia-
tion}. For the properties of BV functions we refer to [19, 23, 11].
A typical example of a function/(x, p) satisfying conditions (1.3) is the nonpara-
metric area integrand yl + \p\2, to which corresponds the functional

(1.4) F(u)= f Vl+|£>«|2+ f H(x)u(x)dx.

Received by the editors January 5, 1984.

1980 Mathematics Subject Classification. Primary 49F10; Secondary 35D99, 26B30.

©1985 American Mathematical Society

0002-9947/85 $1.00 + $.25 per page

483

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 484 GABRIELE ANZELLOTTl

Other examples, where we take H = 0, are the functionals

(1.5) f ,/l + a .(*)/>,.«£>

«,
where the functions atj(x) are continuous and satisfy aij(x) = a,,(x), |£|2 <
«<,(*)«,{, <M|€|2 to all* eK»,

(1.6) f VI+ l¿>"|\

•'o
Í2
A:>1,

(1.7) /)8(|Z)«|), where/?(r)= 1 f°r'<¡>

■'0 12/ - 1 for i > 1,

(1.8) f |Z)„|.
For the area functional A(u) = /ßyl + |Z)w|2, it is well known [17, 24, 14] that
the local minima in BV(ß) are actually real analytic. The regularity C2" of the BV
minima is known also for the functionals of the type (1.4) if the curvature H is
Lipschitz continuous [18, 8, 29], and for the functional (1.5) if the coefficients a,7(x)
are of class C2-" [15]. Actually, the regularity C2a of the BV minima is proved in [15]
for a whole class of functionals satisfying suitable structural conditions (essentially
the conditions under which one is able to prove interior a priori bounds of the
gradient for the smooth solutions of the corresponding Euler equation [22, 15, 26,
16]). In most cases, however (for instance for the functional (1.6)), the regularity of
the local minima in BV is not known and, in many cases, one can easily construct
examples of local minima which are discontinuous along an (n — l)-dimensional
surface. This happens for instance for functionals (1.7) and (1.8), for (1.5) if the
coefficients are not sufficiently smooth and for (1.4) if the curvature H is for
instance only L00. Then it is natural to consider the problem of writing the Euler
equation (1.2) in a suitable weak formulation that still holds for minima u g BV(ß).
We get the following result: if the function f(x, p) satisfies a few natural
hypotheses of differentiability in p (as in the assumptions of Theorem 3.6) and if
u g BV(ß) is a local minimum for F(u) (/n/(x, Du) is defined for BV functions as
in (3.1)), then one has

4-F(u + t<p) = f H(x)<p(x)dx+ f fp(x,(Du)a(x))-(D<p)a(x)dx

dt (=o JQ JQ

<1-9) +/4'-ifi<'>)'^7<*rf
Ja \ \Du\ I \Dtp\
=0
for all admissible test functions <p,i.e. for all <pG BV(ß) such that
(1.10,i) sptçfQ,
(1.10, ii) \D<p( <k \Du(

(i.e. \D<p\s is absolutely continuous with respect to \Du\s),where Du = (Du)" + (Du)s

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 485

is the Lebesgue decomposition of the measure Du in an absolutely continuous and a

singular part with respect to the «-dimensional Lebesgue measure =5?"; we denote by
(Du)"(x) the density of the measure (Du)a with respect to .S"' and by Du/\Du\ the
density of Du with respect to its absolute variation Du. Finally, f°(x,p) is a
positively homogeneous function in p, defined in (2.3), whose existence has to be
assumed and that coincides with the recession function of f(x, ■) in case/is convex
in p. For the notions of absolute continuity, Lebesgue decomposition and density (or
derivative) of a measure À with respect to a positive measure ¡x, we refer to [11, §2.9].
We recall that an equation of the type (1.9) has been introduced in [6] for the
energy functional in Hencky plasticity.
The key point in (1.9) is that the test functions (p are allowed to have "jumps", but
only where the minimum u itself has a "jump". The fact of considering variations
<pg BV(ß) is important, because equation (1.9), if it is considered only for all
functions <p g Co°(ß) (that is for <pg H0u(ß)), does not carry all the information
related to the fact that u is a local minimum (Remark 3.12). On the other hand, the
restriction (1.10,ii) is natural in two ways: first, the derivative (d/dt)F(u + i<p)|,=0
exists if and only if (1.10, ii) holds; second, under natural assumptions on /, if a
function u G BV(ß) verifies equation (1.9) for all the admissible test functions, then
« is a local minimum for F(u) (Theorem 3.10).
The validity of equation (1.9) is also proved, for a restricted class of admissible
variations, if / is not differentiable in p at the points p where / takes the value zero
(Theorem 3.9). This result applies in particular to the case where/(x, p) is positively
homogeneous inp, as for the functional (1.8).
The results on the differentiability of the functional F(u) are obtained from
corresponding results, given in §2, for the functional l(fi) = /n/(x, ft), where ju is a
general Revalued Borel measure in ß. One could use the general results of §2 also to
get an Euler equation for functionals of the type

(1.11) f f(x,e(u)),
where u g Lx(ß,R") is a vector field of bounded deformation, i.e. the distributions
e¡j(u) = 2[D¡uJ + DjU'] are measures of bounded total variation in ß [27]. However,
we shall not develop such a theory, as it is totally similar to the one given in §3 for
F(u).
In §2 we collect also a few simple facts about the extension to the space of
measures of the functional I(r¡) = jf(x, r¡(x)) dx. We shall adopt the approach of
[25, 15, 13, 9]. Recent papers [28, 10] give an extension by making use of convex
analysis and duality theory. In fact, most examples of integrands f(x, p) in the
calculus of variations are convex in p, on the other hand the convexity is not needed
for many of our purposes and we shall assume it only when it is necessary.
We conclude with a few remarks.
First: if « g BV(ß) is a local minimum for the functional (1.1) and we write
equation (1.9) for all <pg C0°°(ß), then we get that the vector field <jd(x) =
fp(x,(Du)"(x)) satisfies di\ip = H and \p belongs to the class X(Q)n= {\p g
L°°(ß)|divi// g L"(ß)} considered in [2]. This fact has been one of the motivations
for the work done in [2].

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 486 GABRIELE ANZELLOTTI

Second: Once we have the Euler equation for the functional F(u), it will be
natural to try a standard technique to get information on the regularity of a local
minimum u of F, i.e. to write equation (1.9) for the test functions of the type y = r\u,
where r¡ g C0°°(ß) (that certainly satisfy condition (1.10)). This has been done
already for the energy functional in Hencky plasticity [3], compare [21], and it will
be done in a forthcoming paper for general functionals of the type (1.1).
Third: For functions in BV(ß) the Euler equation (1.9) suggests the possibility of
a new weak formulation for the quasilinear equations of the type div^4(x, Du) =
B(x), where A(x, p): ß X R" -» R" and B(x): ß -» R, which are not necessarily the
Euler equation of any functional.
I would like to thank G. Dal Maso for some useful conversations about the
semicontinuity results in [9].

2. Directional derivatives of the functional I(ß) = fr¡f(x, ti). First, we are going to
define a real valued Borel measure/(x, it) on ß for all functions/(x, p): ß X R* -> R
that satisfy a few suitable conditions and for all ju,G M($l,Rk), where M(ß,R*)
denotes the space of the Revalued Borel measures in ß.
Then we shall consider the functional 7(/x) = jaf(x, ¡u) and we shall obtain a
formula for the derivative (d/dt)I(¡i + tß)\l=Q under suitable assumptions on/and
ß-
Let us consider the following conditions:
(2.1) f(x,p)>0,
. . /is such that for all Borel measurable functions r\: ß —>R*,
the function x -» f(x, tj(x)) is Borel measurable,
(2.3) for all x g ß there exists the finite limit

limf(x,?-)t=f0(x,p).

We remark that condition (2.2) is satisfied for instance when / is a Borel function
in (x, p). Condition (2.3) says that the function/is asymptotically linear for large/;.
It is immediately evident that the function/°(x, p) is positively homogeneous in
p, i.e.
/°(x, sp) = sf°(x, p) for allx, p and s > 0,
and that/°(x, p) satisfies condition (2.2) if/does.
Now we can give the
Definition 2.1. Let f satisfy conditions (2.1), (2.2) and (2.3). Then for any measure
H G M(ß, Rk) we define a positive measure f(x, ¡u) on ß as

(2.4) ( f(x, a) = / f{x, fix)) dx + f Ax, ^(x)) d\fi\'

JB JB JB \ d\ll\ )
for all Borel sets ficß, where li = ¡i" + lis is the Lebesgue decomposition of the
measure ¡i in an absolutely continuous and a singular part with respect to the Lebesgue
measure Si"1. We denote by li"(x) the density of the measure pf with respect toáf" and
by (dns/d\¡x\s)(x) the density of ¡is with respect to |ju|s.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 487

In formula (2.4) we could write (dp./d\n\)(x) instead of (dns/d\n\s)(x), because

the two functions are equal |/x|s-a.e.
If the function/is such that
(2.5) \f(x,p)\<c(l+\p\),
then one also has |/°(x, p)\ < c\p\ and

(/(x,M)<c(.*"(0) + |m|(Q));
hence, if (2.2), (2.3), (2.5) hold and |jn|(ß), ^"(ß) are finite, then formula (2.4)
defines a measure/(x, ju) also without imposing condition (2.1), and the measure
/(x, p.) has in this case a finite total variation.
We should like to remark that one could easily extend the preceding definitions to
more general measures fi and to other positive measures than ££". We also remark
that for a measure /x = r\(x)££n, which is absolutely continuous with respect toiP",
one has jBf(x, li) = JBf(x, r¡(x)) dx, while for a singular measure n = ¡is we have

//<-'>=//I-ííH*
dV\
du

We shall use the following simple fact.

Lemma 2.2. If f(x, p) is positively homogeneous in p, then for all Borel sets ficß,
one has

<■>
¡A:-t(x))*'-/.'(*•IH*■'+Lf[x'ti(x))*J-
iffix + fi2 = li and Ltx,¡i2 are mutually singular, and
d¡i

if a is a positive measure such that li «: a.

Proof. As ju1and ¡i2 are mutually singular, we have |jti| = \llx\ + \li2\; hence

/ fix, -^-(x)\ d\ß\ -{fix, £-{xj) d\ßx\ +( fix, -^-(x)) d\n2\

JB \ d\fl\ j JB \ d\p\ ! JB \ d\il\ I
and (i) follows because one also has that
dp , , d[ix d¡xx ¿- . ¡\-
—■—(x) =-(x) = —-—(x) \nx\-a.e. in ß,
d\n\ d\n\ i/|/i,|
dP t \ dV-i¡ \ dV-2i ^ ii • r,
—£—(x) =-(x) =-(x) \Lt2\-a.e. in ß.
d\n\ d\fi\ d\n2\
Now, writing ¡i = li" + lis, by (2.4) and statement (i) we get immediately that

= r I (dLt/do)(x) '
hJ\ ' (dWda)(x)} da
and (ii) follows because of the homogeneity of /in p. Q.E.D.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 488 GABRIELE ANZELLOTTI

Now we shall see another way of writing the measure/(x, li) which has a useful
geometrical meaning when ll = Du and u g BV(ß). Let us consider the function
f(x, p, t): QxR'xR'x[0,+oo)^R defined as
'f(x,p/t)t ifr>0,
(2.6) f(x,p,t) =
[f°(x,p) ifr = 0.
One then has
Theorem 2.3. Assume that f satisfies conditions (2.2), (2.3), (2.5) and that
|ix|(ß), JSf"(ß) are finite. Then one has

(2.7) //(*. M)-/B/(*.è(x).^W)rf«.

where a is any positive Borel measure such that |u| + £?" <§: a.

Proof. First, it is easy to see that for any pair of Borel functions r/: ß -* R* and £:
ß —»[0, + oo), the function x -» f(x, tj(x), |(x)) is also Borel measurable; hence the
integral to the right in (2.7) is well defined. Consider then the R'xR valued
measures X = (li, if"), Xx = (li", Sen) and \2 = (<iJ,0). We have that \x + \2 = X
and that Xx, X2 are mutually singular; hence, by Lemma 2.2 we get

Í fLd}L{x) d^_{x)\m
VI d\xy ' d\x\ j '

which proves (2.7) for the particular choice a = |X|. The proof for general a follows
by statement (ii) of Lemma 2.2. Q.E.D.
Now we shall study the differentiability of I(ll). From now on, we shall assume
that / satisfies (2.2), (2.3), (2.5), that if"(ß) < + oo and we shall only consider
measures ju, ß g M(ß, R*) of finite total variation.

Theorem 2.4. Assume that f(x, p) is differentiable in p for all (x, p) G ß X R* and
that f°(x, p) is differentiable in p for all (x, p) G ß X (R* - {0}). Assume also that

(2.8) \fpix, p)\ < M, \fp°(x,p)\^M.

Then the functional I(li) is differentiable at the point li in the direction ß if and only if
\ß\s «: \fi\s and in such a case one has

(2.9) |/(u + tß) = f fp(x,Ll"(x))-ß"(x)dx

t=0 JQ.

i/;U(4flW^
y" I d\p\ I d\ß\

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 489

Proof. Let ll be fixed. For each ß we consider the Lebesgue decomposition

ßs = ßsa _|_ßss 0£ tjje measure ßs in an absolutely continuous and a singular part
with respect to |/x|J. Then we have
ju + tß = n" + ¡Xs + tß" + tßsa + tßss

and, using Lemma 2.2(i) and (ii), we get

(2.10) /(ju + tß)= f fix, paix) + tß"(x)) dx

dfis dß'a
+ ( f°[ x, -^ix) + t^(x)\ d\ix\S+ \t\f /°(x, ß").
4ß\ d\n\
Now we obviously have that

d
jj fix, ll«(x) + tß'ix)) dx\t„0 = I fp(x, ll"(x)) ■ß«(x) dx.

Moreover, as /°(x, p) is differentiable in p for p =*=

0 and \(d¡is/d\Li\s)(x)\ = 1
|ju,|J-a.e., we have

dßsa , ,
+ t~77{x) d\it\
* Jo \ d\p\ d\ii\ r-0

(2.11)
Ja \ d\p\ I d\p\

M'^u)i dtts . \
#-
dßs
d\ßs

and we also have

d+
^(\<\(foi*'ß"))rtC^^r^Wañ^r} r= 0

It is evident that the right and left derivatives in the last formula coincide and are
then zero if and only if faf°(x, ßss) = 0, i.e. if and only if ßss = 0, that is if and
only if ßs <k \ll\s. Q.E.D.
The preceding theorem does not apply when the integrand/(x, p) is positive and
positively homogeneous in/?, because in this case/(x, p) cannot be differentiable in
p at p = 0. Also in this case, however, we can differentiate the functional f(jtt) in a
suitably restricted set of directions. More generally, we have the following result.

Theorem 2.5. Assume that for every fixed x g ß the function f(x, p) is differentia-
ble in p for all p ¥= 0 and assume that f(x, 0) = 0 for all x such that f(x, ■) is not
differentiable at p = 0. Assume also that f°(x,p) is differentiable in p for all
(x, p) G ß x (R* - {0}). Finally assume that

\fp(x,p)\^M, \fp°ix,p)\<M,

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 490 GABRIELE ANZELLOTTI

when the derivatives exist. Then, if ¡i, ß are such that

(2.12,i) r«lM|\
(2.12,ii) j8a(x) = 0 <£n-a.e.inthesetT = {x G ß|jtt"(x) = 0},

the functional l(fi) is differentiable at the point ll in the direction ß and formula (2.9)
still holds.
Proof. We start again from (2.10), where ßss = 0 by (2.12, i). For all x g ß \ T
we have

jtf(x, ua(x) + tß"(x)) = f„(x,LL"(x))-ßa(x).

On the other hand, because of (2.12, ii), for i? "-almost all x g T we have ßa(x) = 0
and

jtf(x,Lt"(x) + tß«(x)) = 0.

Formula (2.9) follows then as in the preceding theorem. Q.E.D.

We remark that conditions (2.12) are equivalent to saying that ß «: \fi\.

3. The Euler equation in BV(ß). In this section we shall consider integrands

/(x, p): ß X R" -> R that, unless it is otherwise specified, satisfy conditions (2.2),
(2.3), (2.5), and we shall apply the results of the preceding section to the functional

(3.1) F(u)= f f(x,Du) = f f(x,(Du)"(x))dx+ f Ax, —(*)) \Du\*

Jn Ja Ja \ \Du\ I
which is defined for all functions u g BV(ß); we assume that ß c R" is bounded.
The functional F(u) is an extension to the space BV(ß) of the functional
fQf(x, Du(x))dx, which is defined elementarily for u g C\ti) n HU1(Q). The
interest for considering this extension stems mainly from a few known facts that we
would like to recall briefly here for both completeness and later reference.
Fact 3.1. Assume that the function f(x, p, t) defined in (2.6) is continuous. Then, if
Uj g BV(ß) and

Uj-*u in L\Q),
(3-2) r I ¡ ¡a r I-1—ä
/ yi+\Du:\ ^ I il+\Du\ forj ^oo,

one also has /n/(x, Duj) -* /n/(x, Du).

Fact 3.2. Assume that f(x, p,t) is continuous, and that it is convex in (p, t) for all
fixed x g ß. Then, ifu, Uj G BV(ß) are such that

(3.3, i) Uj-*u inL\Ü),

(3.3, ii) f ¡Du! < C for allj,

Jo J

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 491

one also has

liminf f f(x, Du) > f f(x, Du).

The proof of Facts 3.1, 3.2 is an immediate consequence of Theorem 2.3 and of
Theorems 3 and 2 of [25], which are given for general functionals of a measure (and
would apply also to the functional (1.11) [5]).
It is worth recalling that under the assumption of convexity in p and (2.5), a
necessary and sufficient condition on/(x, p) in order for/(x, p, t) to be continuous
is that [7] for all x g ß and e > 0 there exists a number 8 > 0 such that

\f(x0,p)-f(x,p)\ <s(l+\p\)
for all p and all x with |x — x0| < 8. We recall also (same proof as in [4, Theorem 1])
that for all the functions u G BV(ß) there exists a sequence u, £ Cx(ü) n BV(ß)
such that (3.2) holds.
It is not difficult to see that Fact 3.2 holds under the weaker assumption that /is
lower-semicontinuous; and we recall that if f(x, p) is lower-semicontinuous, convex
in p and satisfies (2.5), then f(x, p, t) is also lower-semicontinuous. Condition
(3.3, ii) can be dropped if one assumes also that f(x, p) > a\p\ - b for all x G ß,
p G R*, where a > 0, b > 0. As a general reference for these and other more general
semicontinuity results, see [9]. See also the recent paper [30].
In conclusion, if f(x, p, t) is continuous and convex in (p, t) and/(x, p) > a\p\
— b for all x, p, then one has

Ujg cHß)nBV(ß),'
/ fix, Du) = mini liminf / /(x, DuÂx))dx
Jo ;->on •'O Uj —yu in L*(ß)

compare [13, 9].

If we assume that ß has a Lipschitz boundary and that/(x, p) is defined also for
x g 3ß, we can consider the functional [15, 13, 1]

G(u)= f fix, Du) + f f°ix,pQix)[gix) - uix)]) dH"~\

where g g L1(dil) is a given function and va(x) is the outward unit normal to 3ß at
x. Again one easily has (appendix):
Fact 3.3. Assume that the function f(x, p, t) is continuous in ß XR" X [0, + oo).
Then, for all u g BV(ß) and g G L!(3ß) there exists a sequence of functions u} G
Cx(ß) n BV(ß) such that u}\m = g and that
(3.4) uj ^u inL\Q), G(uy) -» G(w).
Fact 3.4. Assume that f(x, p, t) is lower-semicontinuous on ß X R" X [0, + oo) and
is convex in (p, t) for each fixed x G ß, and that f(x, p) ^ a\p\ - b for all x, p.
Then, for any fixed g g L1(3ß) and for any sequence Uj G BV(ß) such that Uj -> u in
L:(ß) one has
liminfG(wy) > G(u).

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 492 GABRIELE ANZELLOTTI

By Facts 3.3 and 3.4 immediately follows

Fact 3.5. If f(x, p, t) satisfies the assumptions of Facts 3.3 and 3.4, then G(u) is
the greatest functional on BV(ß) which is lower-semicontinuous with respect to the
Ll(Sl)-convergence and satisfies G(u) < fnf(x, Du(x)) dx for all functions u G Cl(û)
n Hll(ü) with u = g on 3ß. Moreover one has

inf G(u)= inf f f(x, Du(x)) dx.

ugBV(H) KeBV(S2)nC'(í2) Jü
u = gon 8fí

The preceding considerations motivate the study of the problem

G(u) -y min, u g BV(ß).
Now we shall study the differentiability and the Euler equation of the functionals
F(u) and G(u). For these purposes we shall not need any convexity and global
continuity assumptions on the integrand/(x, p). Of course, convexity and continu-
ity will be needed in Theorem 3.10.
Theorem 3.6. Assume that f(x, p) is differentiable in p for all x G ß, p G R" and
assume that f°(x, p) is differentiable for ail x g ß, p g R" — {0}. Assume moreover
that

\fp(x,p)\<M, \f°(x, p)\ < M.

Then the functional F(u) is differentiable at the point u g BV(ß) in the direction
<p G BV(ß) if and only if\D<p\s <§: \Du\s, and in such a case one has

(3.5) jtF(u + ttp) = f fp(x,(DuY(x))-(D<f>)a(x)dx

r-0 JQ

Ja \ \Du\ I \D<p\
Theorem 3.6 follows immediately from Theorem 2.4. By Theorem 3.6 we get
Theorem 3.7. Let il be a bounded open set in R" with Lipschitz boundary. Assume
that f(x, p): ß XR" -* R is differentiable in p for all (x, p), that f°(x, p) is
differentiable in p for all (x, p) with p =£0 and that

\fpix,-p)\ < M, \fp°(x,p)\^M

(recall also that we are always assuming (2.2), (2.3), (2.5)). If u G BV(ß) is a
minimum point for the functional G(v) defined in (3.3), then one has
(3.6)
jtG(u + t<p) = f fp(x,(Du)a(x))-(D<p)u(x)dx
r= 0 Ja

hü \ \g(x) —u(x)\
=0

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 493

for all the functions cp G BV(ß) such that

(3.7) \D<p\'4z\Du\',
<p(x) m 0 Hn"l-a.e.inthesetTx = {x g 3ß|M(x) = g(x)}.
Proof. One can use Theorem 3.6 to differentiate faf(x, Du). Because of the
homogeneity of/0, for the boundary term one has
f°(x,va(x)[igix) -«(*)') -upix)])

-Ax. g|*i~1*U(*))l(g(*)- uix))- t9ix)\

\ \g(x)-u(x)\ I
for all x g 3ß where u(x) ¥=g(x), and for sufficiently small t, it follows that

¿/ /°(x, PQ(x)[g(x)-(u(x) + l<f,(x))]) dH-%0

= -{ Ax, g{x) - u{xKa(x)\ s{x) -u{x) 9(X) m*-*

ha \ |g(x)-w(x)¡ j\g(x)-u(x)\

= -( Ax, g(^)-"(x) v (x)\ -VQ(x)eP(x)dH"-1. Q.E.D.

W'\ |g(x)-«(x)| ü J °
Remark 3.8. It is clear that for the differentiability of the boundary term one
needs only the directional derivative of /°(x, p) at each point p ¥= 0 in the radial
direction p, and this derivative always exists because of the homogeneity of/0. A
similar fact happens in the interior of ß in the set N(u) defined in (4.4).
Clearly one can use Theorem 2.5 to get the Euler equation for the case of positive
homogeneous integrands. We have

Theorem 3.9. Assume that f satisfies the assumptions of Theorem 2.5 and let u be a
minium point for the functional G. Then, for all the functions <p= BV(ß) such that

\D<p\ -« |Z?«| ,
(3.8) (£><p)a(x) = 0 2",-a.e.inthesetT= {x(=Q\(Du)a(x) = 0},
(fix) = 0 Hn~l-a.e. in the set Tx = { x g 3ß|M(x) = g(x)},
the equation (3.6) holds.
We shall say that the set of equations (3.6) for all the functions that satisfy (3.7) is
the Euler equation in BV(ß) for the functional G(u). This Euler equation does not
state the vanishing of the first variation of G in all directions <p,nevertheless, we still
have the following result.

Theorem 3.10. Let f(x, p) be as in Theorem 3.7 and assume moreover that
(3.9) fix, p) is convex in p for all fixed x G ß,
(3.10) f(x, p, t) is continuous on ß XR" X [0, + oo).
If a function v G BV(ß) verifies equation (3.6) for all the functions tp G BV(ß) that
satisfy (3.7), then v is a minimum point for the functional G(u) in BV(ß).

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 494 GABRIELE ANZELLOTTI

Proof. Take a sequence of functions v¡ g Cl(ü) n Hl%1(Q)such that i>-|3SÎ= g for

ally, and that converge to v as in (3.4). For any fixed/ set <p-= v¡ — u and consider
the function His): [0,1] -» R defined by //(j) = G(u + ftp,). By (3.9) one has that
f(x, p, t)is convex in (p, t) and it follows that H(s) is a convex function; hence one
also has

(3.11) Givj) = //(l) > //(0) + //(0) = G(«) + 4g(« + upj) Giu),

where we have used (3.6), as <p.satisfies (3.7). By (3.11) and (3.4) the theorem is
proved. Q.E.D.
A condition like (3.10) is needed in the hypotheses of Theorem 3.10, as the
following example shows.
Example 3.11. Consider the open set ß = {xGR|0<x<3} and define a
function/(x, p): ß X R -» R as

' Il +P2 ifx # l,x # 2,

/(x,/>) = /c-i/r+/?2 ifx = i,
(a/l + />2 ifx = 2,

where c is any number such that 3 < c < vTÖ" and 0 < a < c — 3. Take also
g(0) = 0 and g(s) = 1. Then the function v(x) defined as

. , Uc2- 9x ifO < x < 1,

vix) = - -
Uvc2- 9x+(l - \/c2- 9) ifl<x<3
satisfies equation (3.6) for all the functions qp g BV(ß) such that (3.7) holds: in fact,
for all these functions one has G(v) < G(v + <p). On the other hand, v is not a
minimum for G(u), because, if we set for instance
/ x /0 if 0 < x < 2,
wv*)=d if2<x<3,
we have

G(w) = 3 + «<c<c+ -(1-^-9) = G(ü).

Remark 3.12. In Theorem 3.10 we have assumed that v verifies equation (3.6) for
all the variations that satisfy (3.7). To assume that (3.6) is satisfied for all <pg i/u(fl)
would not have been sufficient, in fact, consider the following one-dimensional
example:

ß= {xgR|0<x<2}; fix,p) = Jl +p2; g(0) = 0, g(l) = 1,

/ n /0 ifO < x < 1,
v^ = \l iil<x<2.
It is clear that (Dv)a = 0 and fp(x,(Du)a(x)) = 0 for all x; on the other hand, for
all <pG //¿a(ß) one has (Dy)s = 0 and equation (3.6) is satisfied, while v certainly
does not minimize length.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 495

Clearly, similar results to Theorem 3.10 hold for functionals of type (1.11) and for
local minima.
As a final general remark for this section, we notice that one could study by
similar methods also the subdifferentials of the functionals F(u) and G(u).
4. The functionals/(x, u, Du). Let ß be an open bounded set in R" and consider a
function/(x, j,p):!ixRxR"->R such that
(4.1) \f(x,s,p)\^c(l+\p\),
(4.2) for all (x, s) g ß X R there exists the limit

lim f(x,s,£-)t=f°(x,s,p),

. . for all Borel measurable functions r/: ß -> R" and |: ß -> R,
the function x -* f(x, £(x), r/(x)) is also Borel measurable.
For each function u g BV(ß) we consider a Borel set M(u) such that .SP*'(M(u))
= \Du\siQ \ M(u)) = 0. Following [11, 9] we set
u_(x) = sup{t\{y\u(y) < t) has i? "-density zero at x},
(4.4) " + (*) "* inf {í | { ^ | " ( J7) > t) has <£?"-density zero atx},
N(u) = (x G ß|w + (x) > u.ix)},

and we remark that w_(x), u+(x) are Borel measurable functions on ß. Moreover we
shall consider the measures

Duc = Du\ni«) = ("+(*) - u_ix)) ^Lix)H"-x

\Du\ N(u)

Dud = iDu)s - Duc = Du\M(u)\N(u)

and we shall need the following simple fact (appendix):

Fact 4.1. If<p, u g BV(ß) are such that \Dy\s «: \Du\s, then one has

(4.5,i) \Dtpc\ <^\Duc\,

(4.5,ii) \D<pd\<n\Dud\,
(4.5,iii) H"-l(N(<p)\N(u)) = 0,
(4.5,iv) -^(x) = o(x) —(x) Hnl-a.e.inN(<p),
\Dq>\ \Du\
where a(x) = +1.
Finally, we consider the functional [9]

(4.6) F(u)= [ f(x,u,Du)

Du
= [ fix, u(x),(Du)a(x)) dx+ f Ax, u(x), -^-(x)l \Du\'
Jo Jo \ \Dii\
Du\
Du
+/ r+<vu,,,^(x)
J'N(u)
r\J( ,i\
ds dH'
JJu_(x)
,i t v\ \\ Dili
\Du\

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 496 GABRIELE ANZELLOTTI

and the functional [1]

(4.7) Giu) = Fiu)+f (SiX)f°ix,s,paix))dsdH"-1,

Jaa Ju(x)

where g(x) g L*(3ß) is some fixed function and va is the outward normal to 3ß.
For the functional G(u) we have completely similar results to Facts 3.3, 3.4, 3.5;
the reader himself can provide the statements and proofs. Now we give the Euler
equation for F(u), a similar result holds for G(u).

Theorem 4.2. Assume that:

(i) for all (x, s, p) the function f is differentiable in s and in p, and the function f° is
differentiable in s; moreover one has

\fpix, s, p)\ + \fix, s,p)\+ \fs°ix, s, p)\ < M,

(ii) the function f° is differentiable in p for all (x, s, p) with p + 0, and one has
\fp°(x,s,p)\<M,
(iii) u G BV(ß) is a local minimum for F(u).
Then, for all the functions <p g BV(fi) with spt <p is ß such that \D<p\s «: \Du\s, one
has

(4.8)
d „, x
jtFiu + i<p)
(-0

= ^{fs(x,u(x),(Du)a(x))<p(x)+fp(x,u(x),(Duy(x))(D<py(x)} dx

+í (4-uixl it\(x)hx) +,°(x'u(x)' ñ(x)) w(x))lDuf

+f lfolXíU+(x)í^L{x)\(p+{x)
JN*(u) { \ \Du\ I

-/°(x,«_(x),-^-(x))9.(x))i/i/"-1
\ \Du\ I )

+ ( (/°(x,M + (x),T^-(x))<p_(x)
V(u) I \ \Du\ I

-/°fx,M_(x),T^-(x))<p +(x))j//"-1
\ \Du\ I I
= 0,

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 497

where
Du , . Dtp
N+iu) = x G N(u) C\Ni<p) -
\Du\
(x) - (xfj,
\D<P\
Du Dtp
N~iu)= x G Niu) n Ni<p) (x) (x)
\Du\ \D<p
Proof. Write Fiu + ttp) using formula (4.6). The derivative of the term contain-
ing (Du + tDcpy is easily computed. For the term containing \D(u + t<p)d\, recalling
that|D(u + t<p)d\ «: \Dud\, that one has (Du/\Du[)ix) = (Dud/\Dud\)(x) \Dud\-a.e.,
and using the homogeneity of/° in/?, we get, recalling Lemma 2.2(a),

/ f°lx, u(x) + tcp(x), D(U + t9) (x))\D(u + tcp)ä\

Jn D(u + tw)

f Ax, u(x) + t<p(x),-^-(x) + t-^(x)) |Dw'

Ja \ \Du\ \Du\ )
and the derivative is again easily obtained as in (2.11).
Finally, recalling Fact 4.1, one has
c /•(«+ <?)+ f0l D(u + tcp)
(x) dsdH"1
JN(u + t<p)A« + "P)_ \ \D(u + t(p)\

n-\
= ( ¡iU+"P)+Ax,s,ot(x)^x))dsdH
JN(u) J(u+tq>)_ \ \Du\ I

because N(u + t<p) a N(u) and (u + tcp) + = (u + t<p)_ exactly at those points of
N(u) that do not belong to N(u + t<p).
Now it is easy to see that for a fixed x G N(u) one has at(x) = +1 for all t that
belong to some neighborhood of zero. Moreover for all these / one has
(u + t<p) + = u + + i<pH Pep
(u + r<p)_= u_+ t<p_
* ^00 (x),
|Z>«| \D<p\
(u + t<p) + = u + + ttp_ Du Dtp
if (x) (x)
(u + t<p)_= u_+ t<p+ j
\D<p
and it follows that
d /•(w+ í<p)+ 0 Du
dt I<«+'<p)- f°\x,s,-(x)
\Du\ K
ds

/°(x,M + (x),-^(x))(P + (x)-/0fx,M_(x),T^(x))<p_(x) inN +(u),

\ \Du\ I \ \Du\ I
l/0(^,"+(x),^y(x))<P_(x)-/0(x,M(x),^-(x))^ + (x) inN-(u).

Q.E.D.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 498 GABRIELE ANZELLOTTI

The assumption in Theorem 4.2 that f(x, s, p) is bounded is somewhat too

restrictive, in fact it is not satisfied for example by the functionals of the type
Ja(u))jl + \Du\2 in which case one has fs(x, u, p) = a'(u)\j\ + \p\2. A more
natural assumption would be

(4.9) \f(x,s,p)\<M(l+\p\).
In fact, similarly to what happens for the functionals of growth m > 1, one is still
able to write the Euler equation under assumption (4.9) provided one takes bounded
test functions. More precisely, one has

Theorem 4.3. Assume that:

(i) for all (x, s, p) the function f is differentiable in s and p and the function f° is
differentiable in s; moreover one has

\fp(x,s,p)\^M, \fs(x,s,p)\^M(l+\p\), \f°(x,s,p)\^M\p\,

(ii) the function f° is differentiable in p for all (x, s, p) with p + 0, and one has
\fp°(x,s,p)\<M,
(iii) u g BV(ß) is a local minimum for F(u).
Then, for all the functions <pg BV(ß) such that
_ _ .. .. . .s . .s
spttpeß, Halloo< +00, |Z><p|<s \Du\ ,
one has (4.8).
The proof of Theorem 4.3 is totally similar to the proof of Theorem 4.2.
We notice that similar results to Theorem 3.9 also hold for the functionals
considered in this section.
A general final remark about functionals of type Jaf(x,u, Du) is that the
integrand could be allowed not to be defined or not to satisfy the assumptions of our
theorems in a set E0 a ßx x Rs of zero //"-dimensional measure (see, for example,
[9]). This is a consequence of the general fact that if u g BV(ß) and Bau with
H" \B) = 0, then one has fB \Du\ = 0.
5. Appendix.
Proof of Fact 3.3. Let u, g C^ß) n BV(ß) be a sequence of functions that
converge to u as in (3.2) and such that v¡ = u on 3ß for all y [19]. Then let
w g Cl(Qi) C\ BV(ß) be such that, for each/, one has

wj\»a= S - »laß- Wj(x) = 0 if dist(x, 3ß) < -,

f \Dw\< f \u-g\ dH"1 + -, ( M < -.

Ja JdQ J JQ J

Set u, = Vj + Wj. Obviously, one has w,|an = g for ally and

Uj -* u in L*(ß),

lim f i/l + \Du,\2= Jaf A +|£>m|2

y-»oo Ja v
+ •'aß
f \u-g\ dH"-1.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATION FOR FUNCTIONALS 499

If we consider the Revalued measures lLj, ll on ß defined by

HjiB)= [ Duj, ll(B)= f Du+ f (g - u)vn(x) dH"1

JBnQ. •'Bnü -/ßn3ß
for all Borel sets Bail, and the R" +^valued measures
otjiB) = (m,(5), ViB)), aiB) = (M(/?), ?'{B)),
then we have
(5.2) aj^a weakly as measures in ß, lim |ö,f(Q) = |a|(ß).
j -*00

Now we notice that

Q{u)~- f fix, a), GiUj)= ( fix,aj)

Ja Ja
and the proof is concluded by Theorem 3 of [25]. Q.E.D.
Proof of Fact 3.4. Let ßj be some open ball containing ß and consider the
function/*(x, p): Qx X R" -» R defined as

Mx )=lf(x>P) ifxeS,
[Mil +\p\) ifxGßt\ß.
By (2.5) one also has that f*(x, p) is lower-semicontinuous. Let h G Hll(üx) be
such that h\dSl = g and for each function u G BV(ß) consider the function u* G
BV^) defined as
u(x) if X G ß,
U"X' \hix) ifxGßL\ß.

For all u g BV(ß) one has

f f*ix,Du*) = G(u) + M f (1 +\Dh\) = G(u) + const(A)

-'ö1 •'ß,\ö

and the proof is concluded noting that if wy-» u in L'(ß) one also has u* -» u* in
L\üx) and using Fact 3.2 for/* in Slx. Q.E.D.
Proof of Fact 4.1. If \Dy\s ■« |Z)i/|J one has D<p = h\Du\s, where /i(x) =
((D<py/\Du\s)(x), and it follows that D<pc= (h\Du\s)\N(ip) = h\(Du)\mJ so that
(4.5, i) is proved.
Now, if we set Nk(<p) = {x g JV(<p)|<p+(x) - <p_(x) > \/k) for k a positive
integer, we have N(<p)\N(u) = \JkNk(tp)\N(u) and, if we assume by contradic-
tion that H"-l(N(<p)\N(u)) > 0, we get

\D<p\iNki<p)\Niu)) > ^H"-liNki<p)\Niu)) > 0

for some k. This contradicts (4.5, i) and proves (4.5, iii).

Then, if \Dud\(B) = 0 for some Borel set B a ß, one has \Du\s(B\ N(u)) = 0
and, by the assumption that \Dq>\s<k \Du\\ it follows that \Dq>\s(B\N(u)) = 0;
hence we get \Dtp\d(B\N(u)) = 0. On the other hand one has \D<pd\(N(u)) =
\D<p\(N(u)\N(<p)) = 0 and (4.5, ii) also follows.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 500 GABRIELE ANZELLOTTI

Finally, we recall that [11, 4.5.9], N(u) and N(q>) are (H"~\ n - l)-countably
rectifiable sets, so that [11, 3.2.19] there exists a Borel set E0 such that H"l(E0) = 0
and such that for all x g N(cp) \ E0 the sets Tan(H"~lzN(cp), x) and
Tan(H"~1zN(u), x) [11, 3.2.16], are (n - l)-dimensional subspaces of R". On the
other hand, from (4.5, iii) it follows obviously that Tan(HnlzN(tp), x) c
Tan(H"~lzN(u), x) for //"^-almost all x g N(<p); hence the two tangent planes
must coincide for //" ^almost all x ^ N(<p). As (Dtp/\D(p\)(x) and (Du/\Du\)(x)
are normal vectors to Tan(H"~lZN(<p), x) and Tan(H"~lZN(u), x) respectively,
(4.5, iv) is proved. Q.E.D.

References
1. G. Anzellotti, Dirichlet problem and removable singularities for functional with linear growth. Boll.
Un. Mat. Ital. C (5) 28 (1981), 141-159.
2. _, Pairings between measures and bounded functions and compensated compactness, Ann. Mat.
Pura Appl. (4) 135 (1983),293-318.
3._On the extremal stress and displacement in Hencky plasticity. Duke Math. J. 51 (1984),
133-147.
4. G. Anzellotti and M. Giaquinta, Funzioni BV e tracce. Rend. Sem. Mat. Padova60 (1978), 1-21.
5. _, Existence of the displacement field for an elasto-plastic body subject to Hencky's law and Von
Mises yield condition, Manuscripta Math. 32 (1980), 101-136.
6._On the existence of the fields of stresses and displacements for an elasto-perfectly plastic body
in static equilibrium,J. Math. Pures Appl. 61 (1982),219-244.
7. M. Boni, Quasi additivit'a e quasi subadditività nell 'integrale ordinario del calcólo délie variazioni alla
Weierstrass, Rend. Istit. Mat. Univ. Trieste 6 (1974), 1-20.
8. E. Bombieri and E. Giusti, Local estimates for the gradient of nonparametric surfaces of prescribed
mean curvature, Comm. Pure Appl. Math. 26 (1973), 381-394.
9. G. Dal Maso, Integral representation on BV(ß) of T-limits of variational integrals, Manuscripta
Math. 30(1980), 387-416.
10. F. Demengel and R. Temam, Convex function of a measure and applications, Indiana J. Math, (to
appear).
H.H. Fédérer, Geometric measure theory. Springer-Verlag, Berlin and New York, 1969.
12. R. Ferro, Variational functional defined on spaces of BV functions and their dependence on boundary
data, Ann. Mat. Pura Appl. 122 (1979),269-287.
13. _, Integral characterization of functionals defined on spaces of BV functions, Rend. Sem. Mat.
Univ. Padova61 (1979), 177-201.
14. C. Gerhardt, On the regularity of solutions to variational problems in BV(ß), Math. Z. 149 (1976),
281-286.
15. M. Giaquinta, G. Módica and J. Soucek, Functionals with linear growth in the calculus of variations,
Comm. Math. Univ. Carolinae 20 (1979), 143-171.
16. D. Gilbarg and N. S. Trudinger, Elliptic P. D.E. of second order. Springer-Verlag, Berlin and
Heidelberg, 1977.
17. E. Giusti, Superfici cartesiane di area minima, Rend. Sem. Mat. Fis. Milano 40 (1970), 3-21.
18._Boundary value problems for non-parametric surfaces of prescribed mean curvature, Ann.
Scuola Norm. Sup. (4) 3 (1976),501-548.
19._Minimal surfaces and functions of bounded variation. Notes on Pure Math. 10, A.C.T.,
Canberra, 1977. (A revised edition is going to be published by Birkhäuser.)
20. C. Goffman and J. Serrín, Sublinear functions of measures and variational integrals. Duke Mat. J. 31
(1964), 159-178.
21. R. Hardt and D. Kinderlehrer, Elastic plastic deformation, Univ. of Minnesota Math, reports, 1983.
22. O. A. Ladizenskaia and N. N. Ural'tseva, Local estimates for gradients of solutions of non-uniformly
elliptic and parabolic equations, Comm. Pure Appl. Math. 83 (1970), 677-703.
23. U. Massari and M. Miranda, Minimal surfaces of codimension one monograph, Notas de Matemática.
North-Holland, Amsterdam, 1984.
24. M. Miranda, Un principio di massimo forte per le frontière minimali. Rend. Sem. Mat. Padova 45
(1971), 355-356.

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use

 THE EULER EQUATIONS FOR FUNCTIONALS 501

25. Yu. G. Reschetniak, Weak convergence of completely additive vector functions on a set, Sibirsk. Mat.
Z. 9 (1968), 1386-1394. (Translated)
26. L. Simon, Interior gradient bounds for non-uniformly elliptic equations, Indiana Univ. Math. J. 25
(1976), 821-855.
27. G. Strang and R. Temam, Functions of bounded deformation, Arch. Rational Mech. Anal. 75 (1980),
7-21.
28. R. Temam, Approximation de fonction convexes sur un espace de mesures et applications, Canad.
Math. Bull. (4) 25 (1982), 392-413.
29. N. S. Trudinger, Gradient estimates and mean curvature, Math. Z. 131 (1973), 165-175.
30. E. De Giorgi, G. Buttazzo and G. Dal Maso, On the tower semicontinuity of certain integral
functionals. Rend. Accad. Naz. Lincei (1984).

DlPARTIMENTO DI MATEMÁTICA,UlMIVERSITÀDI TRENTO, 38050 POVO (TN), ITALIA

License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use