12 CHAPTER 1 Introduction: Maxwell Equations
densi. .ty .qp due21 to this. change in localization,, let us treat it as a mathematical
&strzbutzon, that is, as the mapping ~ --~ Y qp ~, where ~ denotes a
so-called "test function". Expanding ~ to first order and integrating by
parts, we have
qp ~/= ~ q+(x) [~(x + u(x) / 2) - ~(x - u(x) / 2)] dx
~ J q+ u. grad ~ = J-div(q÷ u) ~,
hence qp = -div p, where p = q+ u. This field p, soon to be identified
with the one in (3), is the polarization of the dielectric.
Exercise 1.7. Try to do the same computation "the other way around",
by starting from ~ qp ~ = ~ q+(x - u(x)/2) ~(x) - ~..., etc. Why does it go
wrong this way?
The macroscopic manifestation of this local charge splitting is thus
the appearance of a distribution of charges in what was initially an
electrically neutral medium. Moreover, if the polarization changes with
time, the motion of charges + and - in opposite directions amounts to
a current density jp = 3t(qu) = 3 tp. (Note that 3 t qp + div jp = 0, as it
should be.)
We might treat this current density on the same footing as j, and
replace the polarized matter by vacuum plus polarization current. Then
d = ¢0 e, and Eqs. (1) and (3) would combine to give
(14)
- 3t(£0 e) +
rot h = j
+ ~)tP"
Instead, we use our option (cf. Exer. 1.1) to charge 3 t p on the account of
Eq. (3), by setting d = g0 e + p, hence - 3td + rot h = -~)t(~0 e + p) + rot h =
j + 3tp - 3 t p = j, leaving e unchanged. This separates macroscopic
21In the theory of distributions [Sc], functions are not defined by their values at points
of their domain of definition, but via their effect on other functions, called test functions. So,
typically, a function f over some domain D is known if one is given all integrals JD f V,
for all smooth ~ supported in D. It is thus allowable to identify f with the linear
mapping ~ --~ ~D f V. (The arrowed notation for maps is discussed in A.1.9.) This has the
advantage of making functions appear as special cases of such linear TEST_FUNCTION --~
REAL_NUMBER mappings, hence a useful generalization of the notion of function: One
calls such maps distributions, provided they satisfy some reasonable continuity requirements.
For instance, the map V ~ ~(a), where a is some point inside D, is a distribution
("Dirac's mass" at point a, denoted 8a). The generalization is genuine, since there is no
function fa such that ~(a) = ~D fa~ for all ~. It is useful, because some theories, such as
Fourier transformation, work much better in this framework. The Fourier tranform of the
constant 1, for instance, is not defined as a function, but makes perfect sense as a distribution:
It's (2~) d/2 times a Dirac mass at the origin,
i.e., (2~) d/2 80,
in spatial dimension d.