Pacific Journal of

Mathematics

SINGULAR LIMITS OF QUASILINEAR HYPERBOLIC

SYSTEMS IN A BOUNDED DOMAIN OF R3 WITH
APPLICATIONS TO MAXWELL’S EQUATIONS
A LBERT M ILANI

Vol. 116, No. 1 November 1985

 PACIFIC JOU RN AL OF MATHEMATICS
Vol 116, N o 1, 1985

SIN G U LAR LIMITS OF QUASI- LINEAR HYPERBOLIC

3
SYSTEMS IN A BOUN DED D OMAIN OF R
WITH APPLICATION S TO MAXWELL'S EQUATION S
ALBERT M ILAN I

We establish a singular perturbation result for quasi- linear hyper-

3
bolic systems in a bounded domain of R , depending on a small parame-
ter. We prove and estimate the rate of convergence, as the parameter
tends to zero, of uniformly stable solutions of the complete system to a
solution of the reduced system. This result is then applied to the study of
the convergence of the complete M axwell equations to the quasi- sta-
tionary ones.

1. Introduction.In this paper we are concerned with the singular

perturbation problem consisting in the study of the behavior of the
solution of an initial- boundary value problem for a quasi- linear hyper-
bolic system of the type

(cs) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
ε u tΐ + u
σ t + L(u)u = 0

when the small parameter ε > 0 tends to zero. There are two main
questions which are related: (A) whether the solutions of (cs), which we
denote by uε , tend in some sense to a solution of the parabolic system

(qs) out + L(u)u = 0,

and (B) to compare in some sense the solutions of (cs) to any solution of
(qs) that can be obtained independently.
Singular perturbation problems of this type are extensively considered
in Lions' book [4], in which however the above type of "hyperbolic - »
parabolic" convergence is studied only in the linear case. We present here
some results for the quasi- linear case in a rather special situation, consid-
ering three- dimensional vectors u = u(x,t) defined on a bounded region
Ω x[ 0, + oo[ of R 3+ 1 (although the method of proof suggests that a
proper extension would provide results in any number of dimensions). A
more stringent limitation is that the coefficients of the elliptic operator
L{u)u depend only on the spatial derivatives of u; suitable dependence on
u and ut could be allowed, although it seems possible to obtain ap-
propriate results only if this dependence is on e1/ 2ut rather than on ur

in
112 ALBERT MILANI

The results we obtain are local in time, and we make essential use of
both the strong ellipticity ofzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHG
L(u) and of the presence in (cs) of the
"positive" dissipation term uσ t (we mention in passing that such a term is
essential also in investigating the long time behavior of the solutions of
(cs); see [7]). We are motivated in our study by the "approximation"
problem that arises when displacement currents are neglected in the
determination of the electromagnetic field in a ferromagnetic material.
Such neglect, which is usual for instance in the study of transformer cores,
has the effect of reducing the complete system CS of Maxwell's equations,
which is of hyperbolic type, to the parabolic type system of the quasi- sta-
tionary equations QS. These systems are quasi- linear, because of the
nonlinearity of the magnetic characteristic in ferromagnetic materials; and
it is precisely in this case that displacement currents are known to be
negligible. A first step in studying the related singular perturbation
problem was taken in [6], where we provided results in all of R 3 showing
that indeed the reduced equations QS are the singular limit of the
complete equations CS at the vanishing of the dielectric constant ε.
However, since the quasilinear nature of the equations is due to the
electromagnetic field in a ferromagnetic material, it is physically more
relevant to obtain the same results in a bounded region of space. This in
general presents more difficulties, both because fewer results are available
for the initial- boundary value problem for quasi- linear hyperbolic sys-
tems, and because the (rather standard) technique that is used to obtain
the necessary stability estimates on the solutions of CS in the whole space
would not be effective in a bounded domain, due to the presence of
boundary conditions and the prescription of compatibility conditions on
the initial data that make direct differentiation of the equations with
respect to the space variables of no use. The aim of this work is to show
how, using only differentiation with respect to time, it is possible to
establish for the bounded domain case essentially the same type of
estimate and results that were obtained in [6] for the whole space case.
Differently than in [6], we shall not consider Maxwell's equations directly,
which are a first order system, but rather transform them into a second
order system of the type (cs), using scalar and vector potentials for the
fields, with a procedure already followed in [8]. We conclude by mention-
ing that when the magnetic characteristic is linear, Lions' techniques can
be applied directly to the Maxwell equations; an illustration of this can be
found in [9].

We would like to express our gratitude to Professor Murray Protter

for his kindness and very helpful suggestions.
 SINGULAR LIMITS OF QUASI- LINEAR HYPERBOLIC SYSTEMS 113

3
2. Let Ω c R be a bounded open
The singular limit problem. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQ
domain with smooth boundary 3Ω . We consider the following initial-
uε = uε (x, ί ) e R 3 , x 6 ί l , / > 0:
boundary value problem forzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJ

(2.1) ou[ -
in Ω ,
(2.2) « e ( 0) = zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK
uOε , u'M = ulε
(2.3) «εl 9 Ω = 0
where ε and σ are positive constants, u' = du/ dt, djU = {9 M , fyw2, 3yt/ 3}, /
/
1

djUh = duh/ dxj etc.; the α / y's are 3 X 3 real valued matrices satisfying
t
aiJ = β i 7 and du represents the collection of all first order spatial deriva-
tives of u. We assume that for all (sufficiently regular) vector functions
p(x, t) e R 3, the (linear) operator
3
L(p)u= - Σ fl, y (p)9, 9/ w

is uniformly strongly elliptic in the sense that the following matrix

inequality holds:
+ 22
(2.4) 5λ ; Vλ y > λ o | λ| ;

we also assume that the matrices α / y are at least three times continuously
differentiable, with uniformly bounded derivatives, so that L{p), consid-
ered for each fixed p as a linear operator between two Banach spaces, has
uniformly bounded derivatives:
(2.5) Vp, \\L^(p)\\<δ k, 0<k<3,
the norm being that of the proper spaces of linear functional (this
assumption might be somewhat relaxed, assuming instead for L{p) a
polynomial growth of suitable degree). We have explicitly noted in (H ε )
the dependence of the unknown u on the parameter ε, since we want to
compare this system with the reduced parabolic system

(2.6) u
σ '- atj(du)dtdju = 0
in Ω ,
(2.7)
(2.8) =0

formally obtained from (H ε ) by setting ε = 0, and more precisely to

investigate the behavior of solutions of (H ε ) as ε - * 0 and the problem of
114 ALBERT MILAN I

their convergence to the solution of ( P ) . Adapting a definition of Hop-

pensteadt (see [2]), we shall say

(H ε ) degenerates regularly to (P) on [0, T] as ε - > 0 if

D EF IN ITION .zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
f
uQε - > u0 uniformly in Ω and uε - * u, u'ε - » u uniformly on [r, T] X Ω
VT e ]0, JΠ], where uε and w are solutions respectively of (H ε ) and ( P ) on

This notion is necessary in that convergence, if it occurs, is in general

singular in time, due to the loss of one initial condition: we are actually in
the presence of a boundary layer problem in time.
Local existence in time and uniqueness results for (H ε ) can be
established for fixed ε using Kato's theory in [3]; such results however
would not provide uniform estimates with respect to ε, and the (small)
time interval of existence might indeed shrink to 0 as ε - > 0. We need
therefore at first to ensure the existence of a family of solutions of (H ε )
that are stable with respect to the parameter ε, in the sense that such
solutions are all defined on a fixed (small) time interval; and then we must
provide uniform bounds on appropriate norms of the solutions, which will
imply the convergence of (sequences of) such solutions to a limit that is a
solution of (P). These bounds will also provide information relative to the
"approximation" problem, permitting us to estimate the error that is
made in considering (P) instead of (H ε ), with a suitable choice of M 0,
essentially in terms of ε and the difference ulε — u'(0) that accounts for
the loss in initial data. As a final remark we mention that although we
have considered homogeneous systems, the same results would apply in
the more general case, at least if the inhomogeneous terms are small
enough; this would in particular permit us to consider the problem in
non- simply connected domains (see for instancezyxwvutsrqponmlkjihgfedcbaZYXWV
[10] for the parabolic case
of the quasistationary Maxwell equations).
3. Basic assumptions and main results. We recall that in the theory
of quasi- linear hyperbolic systems (see for instance Kato, [3], and its
references), solutions for a second order system such as (H ε ) are usually
s+ι ι s
sought in the space C(0, Ύ \ H ) Π C (0, T: H ) with s > 1 + n/ 2; here
n = 3 and we choose s = 3. We consider therefore the spaces
2 2 3
# 0 = L (Ω ) = ( L ( Ω ) ) ,
m 1 m m 3
//m = H ( Ω ) n H 0( Ω ) , H ( Ω ) = (H (Ω )) for 1 < m < 4,
m
and note || \ \ m the H norm in H m. It is well known that in order for (H ε )
to be solvable, a necessary condition is that the initial data (2.2) satisfy
 SINGULAR LIMITS OF QUASI- LINEAR HYPERBOLIC SYSTEMS 115

certain restrictions, called the compatibility conditions. To describe these,

we start with defining the operatorzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK
M = M(ε, σ) by

w
σ n;
where wn is a given sequence. Then, given a pair of smooth vector
functions {w^v^}, we generate a finite sequence w7 = ^ ( H ^ H ^ ) by
setting
WO
= M ;
O; ^ = ^ 5

w2 = - ε ~ ι {L(dwQ)w0 + w
σ λ );
1 /
w3 = - ε~ {L (3vv0)(9w1)w0 + £ ( 9 ^ ) ^ + σw2 };
1 //
w4 = - ε" {L (3w0 )(3w1 , 3w1 )w0 + L'(dwo)(dw2)wo

+ 2L'(dwo)(dwι )wι + L(3>vo)>v2 + σw3}.

Defining then the set
foτ j < k, 0 < y + k < 4},
we require that the initial data (2.2) satisfy the regularity, compatibility
and boundedness conditions
( 3. 1) M Oe ,W le €= Φ ,
1/ 2
(3.2) K J | 4 4- ε K || 3 = O(l) a sε^ O .

REMARK 1. In [6] the stronger assumption ||M || = 0(1) was made, U 3

but this turns out to be unnecessary. If such is assumed, however, it can

be shown that stronger results would follow.

We are now ready to state our main results. U nder the assumptions
(2.5), (2.6), (3.1) and (3.2) we claim:

THEOREM 1. There exist positive numbers T o and Δ o independent of e

such that all problems (H ε ) have solutions uε Ξ Π y= 0 C y(0, Γo; H 4_j) satisfy-
ing the uniform bounds

(3.3) W e [0, Γo], IM0II4 + 4<(t)\\l+f \\Φ ) \\lds < Δ20,

(3.4) Vί e [0, T Q], \ \ u'ε (ή\ \ 2

2
< Δ20(l + Δ 0 ) 4 ( l + e " ^ - ' / ' ) ,
where 0 < a < 2σ.
116 ALBERT MILANI

TH EOR EM 2.zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
There exist a subsequence {uε } and a vector function u
such that uε - > u in L°°(0, Γo ; H 4) weak* and C(0, Γo ; H 3) uniformly;
u'ε - * u' in L 2 (0, T o; H 3) weak and u is the unique solution of (P) with
u0 = w- lim ε_> 0 uOε in H 4.

(H ere and in the sequel, w- lim and s- lim mean limits taken respec-
tively in the weak and in the strong topologies). We note from Theorem 2
that we do not have uniform convergence of uε to u'\ this singularity is to
be expected, due to the loss of the initial condition on «'. Indeed, from
(2.6) we deduce that the solution u of (P) determined by Theorem 2
2
satisfies u" e L (0, Γo ; H λ ) 9 so that u' e C(0, T O; H 2) (and therefore
u e C(0, Γ o ; H 4), see Remark 2); however, there is not, in general, any
relation between w'(0) and the wlε, and even if \ \ ulε \ \ 3 = 0(1), so that there
exists uλ e H 3 such that uλ = w- lim ε_> 0 ulε in H 3 and ux = s- lim ε _ 0 ulε in
H 2 (by compactness), it need not to be true that uγ = w'(0), unless ulε and
uOε satisfy additional restrictions. These are the so called initialization
conditions; in the present case a sufficient one is the requirement that

(3.5) ||σwlε + L(3wO ε )wO ε ||o - > 0 as ε - * 0;

in fact if it so happens, we have that, since u0 = s- lim ε_^0 uOε in /f3,

L(duo)uo = s- lim ε _ ^ o L(3wO ε )wO ε in / ί 1 , and from

! - uσ '(0)\ \ 0 < ||σ W l - σwlε ||o + ||σwlε

o ) wo - L(duOε )uOε \ \ o
the fact that ux = wr(0). We do not require here any such initialization
condition (we remark that (3.5), together with (3.2), implies that ||wl ε ||0 =
0( 1) , which is in itself one type of initialization condition); indeed, we
have a boundary layer in time. Estimates on the differences uε — u and
uε - u' are provided by

TH EOREM3. Suppose u is a solution of(P) with uQ given in H 4 such that

u e C(0, T; H 4) anduf e L 2 (0, T; H 3)for T < T o. Then Vη > 0

sup k(o- "(oιi4- η

(3.6)zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCB
t*[ o,τ]

This Theorem allows us to estimate the rate of convergence of uε and

uε if ||wOε — u\ \ - » 0 as ε - > 0; moreover, as a consequence for the boundary
 SINGULAR LIMITS OF QUASI- LINEAR HYPERBOLIC SYSTEMS 117

layer problem, we have

1/ 2
Suppose in addition that ||wOε — UQ\ \ X = O(e ) and
COROLLARY.zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
\ \ uu\ \ 0 = 0(1) as e - > 0. Then (H ε ) degenerates regularly to (P) on [0, T] as
ε - > 0.

REMARK 2. Let u be the solution of (P) determined by Theorem 2:

then, as we have already mentioned, ur e C(0, T o; H 2) and therefore
L(du)u e C(0, Γo; H 2). Since w is known to be in C(0, Γo; 7f3), we can
regard W e [0, Γo] L(9w(r)) as a linear strongly elliptic operator with
coefficients in H 2 (Ω ). It has been shown in [5] that this is sufficient to
conclude that Vί u(t) e H 4 (in fact, the Sobolev index 2 is greater than
n/ 2 = 3/ 2 here), so that u e C(0, Γo; i/ 4 ). We point out that this
argument does not depend on the particular values of s and n considered
here, and the condition s > 1 + n/ 2 is sufficient to conclude in the same
way that u e C(0, Γo; H s+1) if wr e L 2(0, Γo; fl,).

REMARK 3. An estimate analogous to (3.6) could be obtained for

7) = 0 (see Remark 5 in §6).
4. Proof of Theorem 1. The proof of Theorem 1 is established with
a standard fixed point technique, carrying out suitable a priori estimates
on the solutions of the linear systems obtained by linearization of (H ε ).
More precisely, consider, for T > 0 the space

Sτ =
7= 0
and for φ e S τ the weighted norm

- sup

•Ό

where we have defined the linear operator M = M(ε, σ, 9,) by

M / = ε / ' + σ/ .
Theorem 1 is proved by a contraction argument on the subspace
= «o.,/ '(O) = uu; [/ ]r 0 < Δ o ,

ll2 ^ Δ 0 ( l + Δ 0 ) 2 ( l
118 ALBERT MILAN I

wherezyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
T o > 0 and Δ o > 0 are to be determined: more precisely, for φ G X
let ψ = θφ and define u = j/ ( φ ) to be the solution of the linear problem
leu" + σw' + L(ψ) u =0
( L O f) |«( 0 ) = «Oe , «'(0) = MiP in Ω ,

We claim that

P R OP OSI TI ON 1. There exist T o > 0 and Δ o > 0 independent of e such

that
(4.1) sfmaps Xτ Δ /wto itself;

(4.2) s/ is a contraction in Xτ Λ for the norm

{
re[0, Γ 0 ]

The conclusion of Theorem 1 would then follow easily from Proposi-

tion 1.

REMARK 4. Because of (3.1), the initial data in (L O ε ) satisfy the

regularity and compatibility conditions necessary in order that the system
have a solution u G 5 Γ . This can be shown by a direct application of
Kato's result in §2.6 of [3]; the reservation made there on the regularity
result that is needed for the operator L( ψ ) has been eliminated in the
already mentioned paper [5]. Before we proceed to the proof of Proposi-
tion 1, it is useful to recall the integration by parts formulas we shall use
in the sequel. N oting by ( , ) the ordinary inner product in L 2(Ω ), we
have:

LEMMA 1. V/? G Cι φ ) , Vw G H 2 ( Ω ) , Vt; G H V

u du
(4.3) (L(p)u9υ ) = {Σ <*,j(p)dι >djv) - ^ ( P ^ P ' * > v)>

where {f,g}*- >ζo(p,'d\ f,g) is for all fixed p a bilinear continuous form
2
on L (Ω ) such that, because of (2.5), for suitable M(p) and M > 0
(4.4)
V/ Vg e L 2 (Ω ), \ UP, dp f, g)\ < Aί(/ ;)||/ l|o|g
(here and in the sequel we write | \ mfor the norm in C m ( Ω ) ) .
 SINGULAR LIMITS OF QUASI- LINEAR HYPERBOLIC SYSTEMS 119

LEMMA 2. V/> e C°(0, Γ C^Ω )) n C^O, Γ;C°(Ω )), Vw e

2 2
L (0, Γ; H (Ω )) Π JϊHO, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDC
T; H λ \

(4.5) ( L( ^) u , II') = \ I Σ ( ^ ( ^ ) θ , n , 9,«) - \ ξ τ (p9 p'\ duf θu)

w/zere {/ , g} - > {x(/ ?, / ?'; / , g) has analogous properties as £ 0 ; (4.5) follows

immediately from (4.3), recalling that atj = ^ 7 / .
X
We define then for fixed/ ? the quadratic forms on H (Ω )

X 2
and the bilinear form on H (Ω ) X L (Ω )
β 3 ( Λg ) = ^ o ( ^ , 9p ; 9Λg)
and notice that because of (2.4), (2.5) and Poincare's inequality, Qx is an
equivalent norm on H x and Q2 and ^3 satisfy
(4.6) \ Q2{u)\ < φ'lollwlli,

(4.7) |β3(/ ,ί)l^cbli||/ llillβllo.

5. Proof of Proposition 1. For φ e Z, set ψ = 3φ and differentiate
(L Oε ) with respect to time (this procedure is rather formal, since u is in
general not regular enough to make sense out of all the differentiations;
because of the linearity of the system however, full justification of the
procedure could be given considering instead a regularization of u in time,
for instance by means of the Friedrichs mollifier). We obtain for 0 < k < 3
the sequences of systems
2+h 1+h h
είu + uσ + L(ψ) u = Gh
in Ω ,

where u = dthu and Gh = L{ψ) u h — dthL(ψ) u.

h
By a linear combination
of these and the linearity of M we also have
2+h +
ι h h

ί
εMu +M
σ u + L{ψ) Mu = MGh
1+h
Mu»(0) = Muhε , Mu (0) = Mu1+htε in Ω ,
120 ALBERT MILAN I

which we shall consider only forzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHG

h = 1,2. Because of the uniform elliptic-
ity of L(ψ) and Poincare's inequality we have

||4- + 2< »'{||M «'||/ + H |i}, 7 = 0,1,2,

2
(5.1) \ \ Mu'\ \ 2 < v{\ \ M u"\ \ o + 11^110 + I | M M ' | | I },

llH lk} 7 = 0,1,

so that we are led to consider the "energy" norms
2 2
= {\ \ u\ \ l + ε\ \ u'\ \ l + \ \ Mu'\ \ l + \ \ M u"\ \ l + ε\ \ Mu"\ \ l},
S {u) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

We proceed now to carry out suitable a priori estimates for such norms
from (L hε ) and (L*ε ). We have from (L O ε ), times uf and recalling (4.5):

(5.2) k||o + Qι( u i/1|o = Q2(u) 4- Q3(u, u')\

from (L*ε ), times Mu":

+
(5.3) ί {**<"»»
r/
) + 2(MG19 Mu")
and finally from (L*J> times M 2u":
d /Ί, , ^ .., , 2

(5.4) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
Λ
fM") + Q3{Mu", M 2u") + 2(MG2, M U"). 2

F rom (5.2), (5.3) and (5.4) we get

= {β 2 ( w) + β 2 ( ^ 0 + eβ 2 (M iι / / )}
^ 5 5^ + {β 3 (w, w') + Qz{Mu\ Mu") + ( ? 3 ( M t Λ M 2 w/ r )}
+ 2( M G l 5 M M ") + 2(M G 2 , M 2U")
— A A. A A. A
Λ τ/ 1 τ/ 1
~ 1 2 3

with the obvious definition of Λ^ and ^4y; we remark that since Qλ is an

r
equivalent norm on H l9 the $• and the j \ ] are equivalent. We claim now
that the right side of (5.5) satisfies the estimate

(5.6) ([ψ]
 SINGULAR LIMITS OF QUASI- LINEAR HYPERBOLIC SYSTEMS 121

θ > 0 is arbitrary, cθ > 0 is independent of ε and λ(Δ) = (1 + Δ) 8 .

wherezyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
At first we have, recalling (4.6) and (4.7):

(5.7) Mxi <

(5.8) \ A2\ <

x(\ \ u%+ \ \ Mu%+ \ \ Mu%)

To estimate A3 we need to estimate the Gj and their combinations MGj.

This is done recalling (2.5), resorting again to Sobolev's imbedding theo-
rems and the elliptic estimates (5.1); we need only to remark that the
linearity of the derivatives of L(ψ) permits us to express suitably the
linear combinations MGj in terms of the combinations Aίψ', M ψ" and
Mu'. We compute

- G2 = L
- G 3 = L '"

so that by opportune linear combinations we obtain

- MGX = eL"(ψ)(ψ', ψ')« + L'(ψ) (Mψ')u + 2eL'(ψ)(ψ')«'.

« + 3L'(ψ)(Aίψ')«'
')«' ~ 2σL"(ψ)(ψ', ψ')«
Because of (2.5) and the Sobolev' imbedding theorems we have then

+ HΨΊIi)ll«Ίl
122 ALBERT MILANI

Combining these with the elliptic estimates (5.1) we arrive after a rather
lengthy but straightforward computation at

ι
cj \ \ MG2\ \ 0

Recalling the bounds implied on ψ = 9φ by the choice φ e l w e have

therefore, writingzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
T and Δ instead of Γo and Δ o , and setting Λ(Δ) =
(1 + Δ) 2 :
2
2(MGlt Mu") < cβ Δ (\ \ λ fu'\ \ l + ||w|i) + θ\ \ Mu"\ \ l

2
, M u") < A(Δ){|ψ'||2Λ(Δ)

+ θ\ \ u'\ \ l + θ\ \ Mu"\ \ l
with arbitrary θ > 0 and cθ determined accordingly. These inequalities,
together with (5.7) and (5.8) permit us to deduce (5.6) easily. Taking θ so
 SINGULAR LIMITS OF QUASI- LINEAR HYPERBOLIC SYSTEMS 123

small that zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

a = 2σ - θ > 0, we have then from (5.5) and (5.6) that
(5.9)

from which we get at first by GronwalΓs inequality

(5.10) ΛΊ 2(«(/ )) < ?Λ (u(0)) exp ceλ(Δ)ζ([<p] 2 + \\φ f 2).

Since φ e l w e have
4 2
ζ tft + a~\ l + Δ) Δ (1 - *- '/ • ) - «0(ί, Δ);
moreover the uniform bounds (3.2) on the initial data are sufficient to
ensure that ^Ki(w(0)) < k, independently of ε. Therefore, setting

we have from (5.10)

(5.11) ^ι{u{t)) < k 2 exp α(Γ, Δ).
Let now m be a sufficiently large integer, and choose at first Δ = Δ o such
that Δ o > k/ m, and subsequently T = T o so that

α ( Γ 0 , Δ 0 ) < l n ( Δ 0 Λm ) 2
(we observe that α(2Γ, Δ o ) is an increasing function of T such that
α(0, Δ o ) = 0); from (5.11) we have then that
^ ( */ ( 0 ) < Δ 0 / m V/ e [ 0 , Γ 0 ] ,
and since from (5.1) we have for all t e [0, T o]
[u(ί)]i ^ cΔ 2 0 ^(u(/ )) ^ cΔ 2 0 ^i( «( / ) ) ,
we can deduce that
sup [ w( ί ) ] i ^ Δ 0 / 2

if m is chosen large enough. From this inequality and (5.9) it is then

immediate to recover the analogous estimate

and finally, to recover the analogous of estimate (3.4), we observe that

since
124 ALBERT MILAN I

we have from (L O ε ) that for allzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGF

θ >0

settin g again α = 2 σ - 0 > O w e h ave th erefore

whence

from which (choosing possibly a larger value for Δ o )

1 2ε
||^|| 2 < Δ 0 M Δ 0 )( l + ε - / V^ ).
The proof of (4.1) of Proposition 1 can therefore be concluded easily. The
proof of (4.2) is then standard, and actually identical to the one given for
the analogous claim in [6], so that we omit it; we need only to remark that
since the weaker norm || ||w is considered, the boundary condition w|aΩ = 0
is sufficient to perform the integration by parts in the space variables that
is used to obtain (4.2). The proof of Proposition 1, and consequently of
Theorem 1, can therefore be completed.
6. Proof of Theorems 2 and 3. In this section we follow closely the
procedure of §8 of [6].

6.1. Because of (3.3) and (3.4) we have that, as ε - » 0:

uε is bounded in C(0, T; i/ 4 ) ,
uε is bounded in L 2 (0, T\ H 3)
(we have set Γo = T for simplicity). There exist therefore a subsequence,
still denoted uε , and a vector function w such that
uε - > w in L°°(0, T; H 4) weak*,
uε - > w' in L 2 (0, T\ H 3) weak,
so that by compactness
uε - > w in C(0, T\ H 3) uniformly
and because of the regularity of L
L(u£)uε - + L(w)w inC(0, T\ H λ ).
Let now ψ be a smooth function such that ψ(jc, Γ) = 0: from (2.1) we get,
upon integrating by parts
I {—ε(Mg, ψ') + σ(Mg, ψ') + (L(wε )wε , ψ ) j = ε(ψ(0), wlε );
 SIN G U LAR LIMITS OF QU ASI- LIN EAR H YPERBOLIC SYSTEMS 125

lettingzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
ε - > 0 and recalling (3.2) we have then
T
( {σ( w' 9ψ ) + (L(w)w 9ψ ) } = 0,
zyxwvutsrqponmlkjihgfedcbaZYXWV

and because of the arbitrariness of ψ

σv/ + L(w)w = 0.
w is therefore a solution of ( P ) on [0, T], with w0 = w- lim F _> 0 uOε in i/ 4
(which exists because of (3.2)). By standard monotonicity methods, such a
solution is easily seen to be unique.

6.2. Suppose now that (P) has a solution u satisfying the assumptions
of Theorem 3: then it follows from (2.6) that
f 2
u G C (0, T\ H 2), u" G C (0, Γ; 7/ 0) Π L ( 0 ? T\ H λ ).
F rom (2.1) and (2.6) we have
'eu" + uσ ' ε + L(3wf )wf = 0,
m" + σw' + L ( 3 M ) M = eu",

from which, setting wε = uε — u:

(6.1) εwε" + σw; + L(duε )wε =[L(3
We have, recalling (4.5) (with/ ? = duε ):

(6.2)

and since by (3.3) and the regularity of L

2 2

we obtain from (6.2) that for suitable a and b > 0

d_
dt
whence (after possibly renaming the constants)

« o + Hlί + */ o'Kllo
K - "'(o)Ho + I K - «ollϊ + / ' ( i +Il«;!l3)ikeii; +
126 ALBERT MILAN I

from which it is immediate, using GronwalΓs inequality and recalling that

/0ΊKII3 ^ const., to deduce that

(6.3) β||<||o + zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

\\w£ + jH k'llo * k τ [ε \ \u u - u'(0)\\l + \\uOe - u£ + e].

We now have for all η > 0, using well known interpolation inequalities:
1 71 3 3
II l l " / !! llf/

1
/ .T o ί rT O \ 1~ " ? / 3/ /.7" 0
\ η/ 3

/ Ikifi- ,* c, / iKH / KII ,

so that because of (3.3) we get from (6.3)

K- "'(o)llo + ike- "olli

that is (3.6).

REMARK 5. Sharper estimates could be obtained if stronger norms of

w were considered instead of the one in (6.2), and even the case η = 0
could be treated. Such estimates would be established using the same
"elliptic procedure" used in §5.
We conclude by proving the Corollary to Theorem 3 (we remark that
its additional assumptions are not enough to control the difference wε',
unless some initialization conditions such as (3.5) are imposed). However
we have from (6.1):
(6.5) β«, w; ) + σ(w;,<)
= (L(du)u - L(duε )uε - ε u" 9 < ) = (Λε, < ) ;
acting as before, and using interpolation, we have
/
> J 0 < c|k||2 < c|k||i |H |
so that from (3.3) and (6.3) it follows that
||A,||0 = O(e^) asε- 0, •
and we can deduce from (6.5) that, for suitable a and k > 0
 SINGULAR LIMITS OF QUASI- LINEAR HYPERBOLIC SYSTEMS 127

whence

(6.6) IKHo < e - "' / ε |K - a'(O)

Then, since forzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
η >1
3 2
^ IIII Ί II<< ~^ ^/
/ I I / πOί
- η ^ S1KII2 Kilo
we have, recalling (3.4) and (6.6):
|k' || 3 _ η < C(l + E
W2eat/ 2eγ
so that ||wε'||3_ η - > 0 uniformly on [T, Γ] VT <Ξ]0, Γ] . If in particular
7) < 3/ 2, H 3 ~ η (Ω ) ^ (C°(Ω )) 3, so that (7/ ε) degenerates regularly to (P)
as ε - > 0.
7. Application to Maxwell's equations. We consider the following
system of the complete Maxwell's equations:

(cs)

= 0 onθΩ
where π is the outward normal to 3Ω , ε and σ are positive constants and
ζ : R 3 - > R 3 is a nonlinear function. It is well known that (7.2) are
redundent if they are satisfied by the initial data (7.4), and that if
n - Bo = 0 on ΘΩ then the additional boundary condition
(7.6) n- B = 0 onθΩ
can be derived from (7.5) and the second of (7.1). It was shown in [8] that
because of (7.1), (7.2) and (7.6), scalar and vector potentials φ and u can
be determined to satisfy the coupled system of equations

( curl u = B
divwr + εφ ' + σφ = 0
uf + Vφ = - E
in Ω ,

XW= 9Ω
(7.8) ί^ π° ™ >
v }
\ div u = 0
so that (CS) transforms into the second order hyperbolic system

ί
eu" + ou'ε + curl ξ (curl uε ) - V div uε = 0
Wε
(0) = wOε, u'ε (0) = ulε in Ω ,

π X uε = 0, div wε = ^ on ΘΩ ,
128 ALBERT MILAN I

where curlzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
uQε = Bo and ε u lε = - Do. The corresponding quasistationary
equations can be transformed into the parabolic system
βu' + curl ξ (curl u) — V div u = 0

ί n X u = 0,
κ(0) = κ 0 ,

div w = 0 on Ω ,
which is related to the fields B and E by the coupled relations
in Ω ,

B = curl w,
' '' U= - M' + v
These systems are somewhat different from those considered in §2, in that
the first order differential operator div appears in the boundary condi-
tions. It will be shown however that, as a peculiarity of Maxwell's
equations, the divergence of u enjoys the same regularity as u itself.
Indeed, we need to modify the spaces H j in the following way, defining
2 2
H o= [u(Ξ L (Ω ) I div u e L (Ω )},
m m
H m= {u(Ξ H ( Ω ) |divw e # ( Ω ) ; n X u = div u = 0 on 3Ω }
for 1 < m < 4; we recall from [1] that /fm coincides with the space
[u G L (Ω ) I curl u e H ^ ^ Ω ) , div u e 7/ m (Ω );
2

n X u = div w = 0 on 3Ω }
on which the norms
|| curl u\ \ m- i + ||div w||m, ||w||m + ||div u\ \ m
are equivalent. We shall denote | \ m the first of these norms in H m. In [8]
a local existence result in time for (M ε ) in a bounded domain was
established, for fixed ε, adapting Kato's results of [3]; such results how-
ever are not stable with respect to ε. We recall from that paper that the
non linear operator u - > curl ξ (curl u) can be written in explicit fashion
as a quasilinear operator
3
u - > Γ(curl u)u = X) atJ(cuή w)3/ 3yw

with suitable 3 x 3 matrices atj obtained by direct differentiation of ξ

adding to this the additional term —v div u we obtain the operator
3
u - > L(curl u)u = Σ 0,- ;(curl w)3zθ7w.

U nder the assumptions made on ξ as in [8] and [6], that is essentially the
requirements that ξ be a strongly monotone asymptotically linear func-
3
tion, derivative of a convex function F: R - > R whose derivatives up to
 SIN G U LAR LIMITS OF QUASI- LIN EAR H YPERBOLIC SYSTEMS 129

the fifth order at least are uniformly bounded, it was shown in [8] that
'a^ = aJi9 that the operator L(p)u is uniformly strongly elliptic, that the
boundary conditions (7.8) are complementing and that integration by
parts formulas analogous to (4.3) and (4.5) hold. We can therefore apply
Theorems 1, 2 and 3 to (M ε ), provided we can take care of the additional
regularity required of div u.
This is done observing that, as a straightforward computation shows,
di\ T(u)u = 0 because of the symmetry of \ζ u) that follows from the
assumption ζ = dF. Since divL(u)u = —div V div w, we derive from the
linearized form of (M ε ) that v = div u is a solution of the linear problem
with constant coefficients
(
ε υ " + υσ ' - div Vί; = 0,
v(0) = div wOε, v'(0) = div uu in Ω ,

to which classical results apply.

REFERENCES

[1] V. G irault and P. A. Raviart, Finite Element Approximation of the Naυ ier- Stokes
Equations, Lect. N otes M ath. 749; Springer- Verlag, Berlin, 1979.
[2] F . H oppensteadt, Properties of solutions of ordinary differential equations with small
parameters, Comm. Pure Appl. M ath., 24 (1971), 807- 840.
[3] T. Kato, Linear and Quasi- Linear Equations of Evolution of Hyperbolic Type, I n :
Hyperbolicity, II ciclo C I M E (1976); Liguori, N apoli, 1977.
[4] J. L. Lions, Perturbation Singulieres dans les Problemes aux Limites et en Control
Optimal, Lect. N otes M ath. 323; Springer- Verlag, Berlin 1973.
[5] A. M ilani, A regularity result for strongly elliptic systems, Boll. U .M .I., (6) 2- B (1983),
641- 654.
[6] , The quasi- stationary Maxwell equations as singular limit of the completed
equations: the Quasi- Linear Case, J. M ath. Anal. Appl., zyxwvutsrqponmlkjihgfedcbaZYXW
102 (1984).
[7] , On the global existence of solutions to the complete quasi- linear Maxwell
equations, in print, Boll. U n . M at. Ital.
[8] , Local in time existence for the complete Maxwell equations with monotone
characteristic in a bounded domain, (IV), Ann. M at. P ura Appl. CXXXI (1982),
233- 254.
[9] , On a singular perturbation problem for the linear Maxwell equations, Rend.
Sem. Mat. Univ. Polit. Torino, 3 8 / 3 (1980), 99- 110.
[10] A. M ilani and A. N egro, On the quasi- stationary Maxwell equations with monotone
characteristics in a multiply connected domain, J. M ath. Anal. Appl., 8 8 / 1 (1982),
216- 230.

Received April 25, 1983. This work has been performed while the author was visiting the
U niversity of California at Berkeley under grant n o. 203.1.27 of the Italian C.N .R.

U N IVERSITA Ό I TORIN O
VI A C. ALBE R TO 10
1- 10123 T O R I N O , ITALY
 PACIFIC JOURNAL OF MATHEMATICS
EDITORS
DONALD BABBITT (Managing Editor) C. C. MOORE
University of California University of California
Los Angeles, CA 90024 Berkeley, CA 94720
J. DUGUNDJI ARTHUR OGUS
University of Southern California University of California
Los Angeles, CA 90089-1113 Berkeley, CA 94720
R. FINN H U G O ROSSI
Stanford University University of Utah
Stanford, CA 94305 Salt Lake City, UT 84112
HERMANN FLASCHKA H. SAMELSON
University of Arizona Stanford University
Tucson, AZ 85721 Stanford, CA 94305

ASSOCIATE EDITORS
R. ARENS E. F. BECKENBACH B. H. NEUMANN F. WOLF K . YOSHIDA
(1906-1982)

SUPPORTING INSTITUTIONS
UNIVERSITY OF ARIZONA UNIVERSITY OF OREGON
UNIVERSITY OF BRITISH COLUMBIA UNIVERSITY OF SOUTHERN CALIFORNIA
CALIFORNIA INSTITUTE OF TECHNOLOGY STANFORD UNIVERSITY
UNIVERSITY OF CALIFORNIA UNIVERSITY OF HAWAII
MONTANA STATE UNIVERSITY UNIVERSITY OF TOKYO
UNIVERSITY OF NEVADA, RENO UNIVERSITY OF UTAH
NEW MEXICO STATE UNIVERSITY WASHINGTON STATE UNIVERSITY
OREGON STATE UNIVERSITY UNIVERSITY OF WASHINGTON
 Pacific Journal of Mathematics
Vol. 116, No. 1 November, 1985

K. Adachi, Le problème de Lévi pour les fibrés grassmanniens et les variétés

drapeaux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
John MacLeod Ball, Remarks on the paper: “Basic calculus of variations” . . . . 7
John Kelly Beem and Phillip E. Parker, Whitney stability of solvability . . . . 11
Alberto Facchini, Decompositions of algebraically compact modules . . . . . . . . 25
S. S. Khare, Finite group action and equivariant bordism . . . . . . . . . . . . . . . . . . . 39
Horst Leptin, A new kind of eigenfunction expansions on groups . . . . . . . . . . . 45
Pei-Kee Lin, Unconditional bases and fixed points of nonexpansive
mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Charles Livingston, Stably irreducible surfaces in S 4 . . . . . . . . . . . . . . . . . . . . . . 77
Kevin Mor McCrimmon, Nonassociative algebras with scalar involution . . . . 85
Albert Milani, Singular limits of quasilinear hyperbolic systems in a
bounded domain of R3 with applications to Maxwell’s equations . . . . . . . 111
Takemi Mizokami, On M-structures and strongly regularly stratifiable
spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Jesper M. Møller, On the homology of spaces of sections of complex
projective bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Nikolaos S. Papageorgiou, Carathéodory convex integrand operators and
probability theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Robert John Piacenza, Transfer in generalized prestack cohomology . . . . . . . 185
Lance W. Small and Adrian R. Wadsworth, Integrality of subrings of
matrix rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
James Michael Wilson, On the atomic decomposition for Hardy spaces . . . . 201