Macroscopic fields and spatial averaging

Let $a$ denote a characteristic molecular spacing and let $L$ denote the shortest spatial scale of the macroscopic field. Choose a smooth averaging window of width $\ell$ with many constituents inside it, but small variation of the resolved field across it. The useful hierarchy is $a\ll\ell\ll L$. Averaging then suppresses atomic fluctuations without erasing the spatial pattern being studied. Near an atomically thin interface, or when the field changes across one unit cell, such a window may not exist.

Linear averaging commutes with spatial derivatives when the window is translation invariant and sufficiently regular. This explains why the averaged Maxwell equations retain their differential structure. Products behave differently: the average of a product need not equal the product of averages. Microscopic energy, force, and local-field corrections therefore require more care than simply replacing every symbol by its mean. A constitutive parameter fitted to one spatial scale should not automatically be trusted for sharp near fields.

Electric polarization counts dipole moment per volume, not net charge per volume. Neighboring dipoles cancel internally when their density is uniform. Where the polarization varies, incomplete cancellation produces a bound volume charge. A finite uniformly polarized block can therefore have zero bound charge throughout its interior while carrying opposite charges on its faces. This does not mean the material has acquired a net charge: the volume and surface contributions must be counted together.

Define the outward unit normal on a material bordering vacuum. The polarization surface charge equals its normal projection, with its sign fixed by which end of each dipole terminates there. Magnetization similarly describes circulating bound currents, and time-dependent polarization gives an additional bound current. Introducing electric displacement and magnetic intensity packages those contributions so that their Maxwell equations show the chosen free sources explicitly. The free/bound division is a modeling convention; total charge and total current remain the physical sources.

A local isotropic linear dielectric uses a scalar susceptibility relating polarization to the macroscopic electric field. Locality means the response at one point ignores field gradients and fields at other points; isotropy means rotating the field rotates the response identically. Neither property follows from averaging alone. Crystals may need tensors, and conducting nanostructures may need spatially nonlocal response. The microscopic field that acts on one molecule can differ substantially from the averaged field used in a measured susceptibility.

Throughout this module real harmonic fields are the real parts of complex amplitudes multiplied by $e^{-i\omega t}$, with positive angular frequency. Time differentiation therefore multiplies an amplitude by $-i\omega$. A causal material can retain memory, so its frequency-domain susceptibility is generally complex and frequency dependent. For a passive local scalar dielectric, the imaginary part of permittivity is nonnegative at positive frequency. Fixing this convention now prevents later loss terms from accidentally describing amplification.