Cli ord algebras and Maxwell's equations
in Lipschitz domains
Alan McIntosh and Marius Mitrea
April, 1998
Abstract
We present a simple, Cli ord algebra based approach to several key
results in the theory of Maxwell's equations in non-smo oth sub domains
m
of R . Among other things, wegive new pro ofs to the b oundary energy
estimates of Rellichtyp e for Maxwell's equations in Lipschitz domains
from [20] and [10], discuss radiation conditions and the case of variable
wavenumb er.
1 Intro duction.
It has b een long recognized that there are fundamental connections b etween
electromagnetism and Cli ord algebras. Indeed, understanding Maxwell's
equations was part of Cli ord's original motivation.
A more recent trend concerns the treatmentofsuch PDE's in domains
with irregular b oundaries. See [11] for an excellent survey of the state of
the art in this eld up to early 1990's and [13], [18] for the role of Cli ord
algebras in this context.
Following work in the context of smo oth domains ([24], [2], [4], [23]), the
three-dimensional Maxwell system
8
3
>
curl E ik H =0 inR n ;
<
3
curl H + ik E =0 inR n ; (M )
3
>
:
2
n E = f 2 L (@ );
1991 Mathematics Subject Classi cation. Primary 42B20, 30G35; Secondary 78A25.
Key words and phrases. Cli ord algebras, Maxwell's equations, Rellichidentities, radiation
conditions, Hardy spaces, Lipschitz domains. 1
(plus appropriate radiation conditions at in nity) in the complementofa
b ounded Lipschitz domain R has b een solved in [20], while the higher
dimensional version (involving di erential forms)
8
m
>
dE ik H =0 inR n ;
<
(M ) H + ik E = 0 in R n ;
m
>
:
2
n ^ E = f 2 L (@ );
m
(plus suitable radiation conditions) for arbitrary Lipschitz domains R ,
has b een solved in [10] (precise de nitions will b e given shortly). See also
[19], [16] for related developments.
One of the key ingredients of the approachin[20], [10]was establishing
estimates to the e ect that the so-called voltage-to-current map, taking n E
into n H and n ^ E into n _ H , resp ectively, is an isomorphism b etween
appropriate b oundary spaces. Ultimately, this comes down to proving norm
equivalences of the following form
(1) kn E k 2 + khn; H ik 2 kn H k 2 + khn; E ik 2
L (@ ) L (@ ) L (@ ) L (@ )
3
in R and, more generally,
(2) kn ^ E k + kn ^ H k kn _ E k + kn _ H k
2 2 2 2
L (@ ) L (@ ) L (@ ) L (@ )
m
in R . As shown in [20, Theorem 4.1] (for Im k>0) and [10, Theorem
8.5] (for arbitrary k 6= 0), the ab ove estimates lead to the invertibilityof
certain magnetostatic (and electrostatic) b oundary integral op erators which,
in turn, are used to solve(M ), m 3.
m
The aim of this pap er is to shed new light on the basic estimates (1), (2)
and to present an alternative, natural approachtoproving such b oundary
energy estimates, develop ed in the framework of Cli ord algebras. This is
an extension of work in [18] where such an approach has rst b een used
for certain PDE's in Lipschitz domains. Besides its intrinsic merit, the
relevance of this work should b e most apparentforthenumerical treatment
of electro-magnetic scattering problems in non-smo oth domains. For the
smo oth context and di erenttechniques see [8].
The departure p oint for us (whichinfactgoesbacktoMaxwell himself;
it has also b een reinvented by M. Riesz) is to write the Maxwell system
as a single equation which, in fact, expresses the Cli ord analyticityofa
certain Cli ord algebra-valued function. Sp eci cally,ifE and H are the 2
3
electric and magnetic comp onents of an electro-magnetic waveinR , then
the entire Maxwell system is equivalent to the Maxwell-Dirac equation
(3) (D + ke )(H + iE e )= 0:
4 4
3
Here E and H are regarded as Cli ord algebra-valued functions in R ,and
D + ke is a p erturb ed Dirac op erator.
4
This is b oth mathematically convenient and physically relevant. In
fact, the same idea has b een extensively used byphysicists who employ
the Lorentz metric in the four dimensional ( at Minkowskian) space-time
domains in order to write the 3D Maxwell system for E =(E ) and
i i=1;2;3
H =(H ) solely in terms of the electromagnetic eld-strength tensor.
i i=1;2;3
Due to its skew symmetry, the latter de nes the exterior di erential form
(sometimes referred to as a Faradaybi-vector eld)
! := (E dx + E dx + E dx ) dt
1 1 2 2 3 3
+H dx dx + H dx dx + H dx dx
1 2 3 2 3 1 3 1 2
(typically, electrical forces are vectorial while magnetic ones are 2-tensors).
See, e.g., the monographs [5], [31], [9], and [28].
The radiation conditions for E and H can also b e naturally rephrased
in terms of ! .Parenthetically, let us note that it is also p ossible to work
with the non-homogeneous form E + H and write Maxwell's equations in
the form (D + ik )(E + H )= 0 (cf.[23], [10]) but this setup app ears to b e
less physically relevant.
However, the most attractive asp ect of this algebraic context is that it
preserves many of the key features of the classical complex function the-
ory.For us, Hardy spaces of monogenic functions in Lipschitz domains and
Cauchyvanishing formulas will play a basic role in the sequel. Based on
3
these, if R is the complement of a b ounded Lipschitz domain and if
n n e + n e + n e denotes the outward unit normal to @ , we shall
1 1 2 2 3 3
show that for any solution ! = H + iE e of (3) there holds
4
kn! !nk k! k ;
2 2
L (@ ) L (@ )
mo dulo residual terms. The Rellichtyp e estimate (1) follows from this. In
fact, a similar argument applies in the higher dimensional case.
Before commencing the ma jor developments we shall intro duce here some
notation. Call a b ounded domain Lipschitz if its b oundary is lo cally given
by graphs of Lipschitz functions in appropriate rectangular co ordinates. We 3
let d denote the surface measure on @ and set n for the outward unit
normal to @ . For a (p ossibly algebra-valued) function u de ned in , the
nontangential maximal function u is given by u (X ):=supfju(Y )j; Y 2
; jX Y j 2dist(Y; @ )g. Also, the nontangential boundary trace on @
of a function u de ned in is taken as the p ointwise nontangential limit
almost everywhere with resp ect to the surface measure on the b oundary.
Acknow ledgments. This researchwas initiated while MM was visiting AM at
Macquarie University. It is a pleasure to have the opp ortunity to thank here
this institution for its hospitalityaswell as the Australian Research Council
for its supp ort. We also wish to thank Rene Grognard for his interesting
and helpful comments.
2 Cli ord algebra rudiments.
m
Recall that the (complex) Cli ord algebra asso ciated with R endowed with
m
the usual Euclidean metric is the minimal enlargementofR to a unitary
complex algebra A ,which is not generated (as an algebra) byany prop er
m
m m
2 2
subspace of R and such that x = jxj , for any x 2 R . This identity
m
m
readily implies that, if fe g is the standard orthonormal basis in R ,
j
j =1
then
2
= 1 and e e = e e for any j 6= k: e
j k k j
j
m
In particular, we identify the canonical basis fe g from R with the algebraic
j
basis of A .Thus, any element u 2A can b e uniquely representedinthe
m m
form
m
X X
0
(4) u = u e ; u 2 C;
I I I
l=0
jI j=l
where e stands for the pro duct e e :::e if I =(i ;i ;:::;i )(wemake
I i i i 1 2 l
1 2
l
the convention that e := 1). For this multi-index I wecall l the length of I
;
P
0
and denote it by jI j.We shall adopt the convention that indicates that
the sum is p erformed only over strictly increasing multi-indices I .
The Cli ordconjugation on A is de ned as the unique complex-linear
m
involution on A for whiche e = e e = 1 for anymulti-index I . In
m I I I I
P P
particular, if u = u e 2A , thenu = u e . Note that
I I m I I
I I
l(l+1)
2
u =( 1) (5) u; 4
l
for any u 2 .For the complex conjugation on A weset
m
!
c
X X
c
:= (6) u e : u = u e
I I I I
I I
P
We de ne the scalar part of u = u e 2A as u := u , and endow
I I m 0
;
I
c 2 c c
A with the natural Euclidean metric hu; u i = juj := (uu ) =(u u) .
m 0 0
Note that (uv ) =(vu) = hu; vi, for any u; v 2A . Another useful
0 0 m
observation is that
(7) jauj = juaj = jujjaj
m
for any u 2A and a 2 R .We also de ne
m
1
u := (8) fu ug
2
for each u 2A .
m
P
0
l
With u as in (4), de ne u := u e and denote by the range of
l I I
jI j=l
l
: A !A . Elements in will b e referred to as l -vectors or di erential
l m m
forms of degree l . The exterior and interior pro duct of forms are de ned for
1 l
2 , u 2 , resp ectively,by
(9) ^ u := ( u) and _ u := ( u):
l+1 l 1
By linearity, these op erations extend to arbitrary u 2A .
m
Two identities which are going to b e of imp ortance in the sequel are
(10) u = ^ u _ u;
and
l
(11) u =( 1) ( ^ u + _ u) ;
1 l
valid for arbitrary 2 and u 2 .
P
m
@
. Then D = We shall work with the Dirac op erator D := e
j
j =1
@x
j
P
m
@ m
2
e = D ,and D = 4, the negative of the Laplacian in R .For
j
j =1
@x
j
reasons which will b ecome more apparent shortly,we shall nd it useful to
m
embed everything into a larger Cli ord algebra, sayR A A .If
m m+1
for k 2 Cwenowset
(12) ID := D + ke
k m+1
2
2
then ID = 4 + k .Furthermore,
k
c
(13) ID = D + ke = ID and ID = D ke = ID :
m+1 k m+1 k
k
k 5
m
If u is a A -valued function de ned in a region of R ,set
m+1
m
X
@u
+ ke u ID u := e
m+1 k j
@x
j
j =1
and
m
X
@u
e + kue : uID :=
j m+1 k
@x
j
j =1
Call u left (right, two-sided) k -monogenic if ID u (uID ,orbothID u and
k k k
uID ) = 0 in . Note that each comp onentofak -monogenic function is
k
2
annihilated by the Helmholtz op erator 4 + k . A function theory for the
p erturb ed Dirac op erator ID whichisrelevant for us here has b een worked
k
out in [21] and [22].
l
The exterior derivative op erator d acts on a -valued function u by
du := (Du). It extends by linearity to arbitrary A -valued functions.
l+1 m
l
Its formal transp ose, ,isgiven by u := (Du)if u is -valued. Once
l 1
again, extends by linearity to arbitrary A -valued functions. Since D =
m
2 2
d + , it follows that d =0, = 0 and (d + d)= 4,asiswell known.
More detailed accounts on these matters can b e found in [1], [7], [12], [18].
Further references can b e found in [14].
3 A distinguished fundamental solution.
The aim is to construct a convenient, explicit fundamental solution for the
Dirac op erator ID . The departure p oint is deriving an explicit expression
k
m
2
for a fundamental solution of the Helmholtz op erator 4 + k in R . If
k = i, this can b e taken to b e minus the kernel of the Bessel p otential