Boundary-Value Problems in Electrostatics
Here is a problem that sounds impossible. A single point charge q floats a
few centimetres above a large sheet of grounded metal. The charge's own field is simple enough — but
the metal responds: electrons in the conductor rearrange, pile up beneath the charge, and set
up a field of their own. To know the real electric field anywhere you would seem to need to know
exactly how all those countless electrons have arranged themselves, which depends on the field, which
depends on the electrons… a snake swallowing its own tail.
Almost every electrostatics problem worth solving has this flavour: you are told the charges in some
places and the potential on some surfaces (a grounded plate is held at
\varphi = 0; a battery holds a plate at a fixed voltage), and you must find
the potential
\varphi everywhere in between. This is a
boundary-value problem. This page is about the master equation these problems obey,
a theorem that makes them tractable, and a beautiful trick — the method of images —
that solves the grounded-plane puzzle above in a single line, with no snakes.
The master equation: Poisson and Laplace
Everything starts from one of Maxwell's equations, Gauss's law in the form
\nabla\cdot\mathbf{E} = \rho/\varepsilon_0, together with the fact that the
electrostatic field is the (downhill) gradient of a potential,
\mathbf{E} = -\nabla\varphi. Substitute the second into the first and the
field drops out, leaving a single equation for the potential alone:
-
With charge present. The potential obeys
\nabla^2\varphi = -\dfrac{\rho}{\varepsilon_0},
where \nabla^2 = \partial_x^2 + \partial_y^2 + \partial_z^2 is the
Laplacian and \rho(\mathbf{r}) is the charge density.
-
In empty space. Wherever \rho = 0 this collapses to
Laplace's
equation,
\nabla^2\varphi = 0.
So the task "find the field" becomes "solve a single second-order partial differential equation for
one scalar function \varphi." That is a huge simplification — one function
instead of three field components — but a differential equation alone does not pin down a unique
answer. Laplace's equation has infinitely many solutions (any constant, any linear ramp
ax + by + cz, and endless combinations). What selects the
physical one is the extra information at the edges: the boundary conditions.
Two kinds appear again and again. A Dirichlet condition fixes the value of
\varphi itself on each boundary surface (a grounded conductor gives
\varphi = 0; a biased plate gives \varphi = V).
A Neumann condition fixes instead the normal derivative
\partial\varphi/\partial n on the surface — equivalently the surface charge
or the normal field. Give a well-posed boundary condition all the way round a region and the potential
inside is completely determined. Why "completely"? That is the next, load-bearing idea.
The uniqueness theorem: why any trick that works, works
Here is the theorem that turns clever guessing into rigorous physics.
-
Statement. In a volume V with charge density
\rho, a solution of Poisson's equation that matches specified values of
\varphi on the entire boundary S is
the only such solution.
-
Consequence. If, by any means whatsoever — inspired guess, symmetry
argument, a fictitious extra charge — you produce a potential that (i) satisfies
\nabla^2\varphi = -\rho/\varepsilon_0 inside and (ii) takes the right
values on the boundary, then you have found the answer. Full stop. There is no
other.
The proof is short and worth seeing. Suppose two solutions \varphi_1 and
\varphi_2 both satisfy Poisson's equation in V
and agree on the boundary. Their difference
u = \varphi_1 - \varphi_2 then satisfies
\nabla^2 u = 0 inside (the \rho terms cancel) and
u = 0 on the boundary. Now integrate \nabla\cdot(u\nabla u)
over the volume and use the divergence theorem:
\int_V |\nabla u|^2 \, dV = \oint_S u\,(\nabla u)\cdot d\mathbf{a} - \int_V u\,\nabla^2 u\, dV = 0,
because the surface term vanishes (u = 0 on S)
and the last term vanishes (\nabla^2 u = 0). But
|\nabla u|^2 \ge 0 everywhere, and an integral of something non-negative can
only be zero if the thing is zero everywhere. So \nabla u = 0, meaning
u is constant; and since it is zero on the boundary, that constant is zero.
Therefore \varphi_1 = \varphi_2. The solution is unique.
This licence — find a solution any way you like and it is guaranteed correct — is exactly
what makes the next trick legitimate.
The method of images: a mirror charge does the whole job
Back to the opening puzzle: a charge +q at height
d above an infinite grounded plane (take the plane as
z = 0, the charge at (0,0,d)). We need a
potential in the region z > 0 that (i) solves Poisson's equation with
just that one point charge, and (ii) equals zero all over the plane. The genius move: throw
the conductor away entirely and pretend there is a second, image charge
-q sitting at the mirror point (0,0,-d). Reveal
the figure step by step to watch the plane turn into a mirror.
Write down the potential of this pair, valid for z > 0:
\varphi(x,y,z) = \frac{1}{4\pi\varepsilon_0}\left[\frac{q}{\sqrt{x^2 + y^2 + (z-d)^2}} - \frac{q}{\sqrt{x^2 + y^2 + (z+d)^2}}\right].
Now check it against the two requirements. On the plane
(z = 0) the two distances become equal,
\sqrt{x^2+y^2+d^2}, so the bracket is
q - q = 0: the potential vanishes on the whole plane, exactly the grounded
boundary condition. In the region z > 0 the only real
charge is the +q at (0,0,d) — the image sits at
z = -d, outside our region — so \varphi solves
Poisson's equation with precisely the right source. Both boxes ticked. By the uniqueness theorem, this
is the potential above a grounded plane. We solved a hard conductor problem by
writing down the field of two point charges.
What the mirror charge tells us: induced charge, force, energy
With \varphi in hand, everything else follows by differentiation.
Induced surface charge. The charge that gathers on the conductor's surface is
\sigma = -\varepsilon_0\,\partial\varphi/\partial z evaluated at
z = 0. Carrying out the derivative and writing
r = \sqrt{x^2+y^2} for the distance from the foot of the charge,
\sigma(r) = -\frac{q\,d}{2\pi\,(r^2 + d^2)^{3/2}}.
It is negative (opposite charge is drawn up under the +q), sharply peaked
right beneath the charge at \sigma(0) = -q/(2\pi d^2), and falls off with
distance. Integrate it over the whole plane and the total induced charge comes to exactly
-q — the plane carries precisely the image's worth of charge, spread out.
The force on the charge. The real charge is attracted to the induced charge on the
plane. But by the images construction, the field acting on the +q is just
the field of the image -q sitting a distance 2d
away. So the force is a plain Coulomb attraction across 2d, directed toward
the plane:
F = -\frac{1}{4\pi\varepsilon_0}\,\frac{q^2}{(2d)^2} = -\frac{q^2}{16\pi\varepsilon_0\,d^2}.
Worked example — the force. Take
q = 1\ \text{nC} = 1\times 10^{-9}\ \text{C} at
d = 1\ \text{mm} = 1\times 10^{-3}\ \text{m}, with
1/(4\pi\varepsilon_0) \approx 8.99\times 10^9\ \text{N·m}^2/\text{C}^2.
Then
|F| = \frac{(8.99\times 10^9)(1\times 10^{-9})^2}{(2\times 10^{-3})^2} = \frac{8.99\times 10^{-9}}{4\times 10^{-6}} \approx 2.2\times 10^{-3}\ \text{N}.
A couple of millinewtons — tiny, but enough to hold a speck of dust against a charged surface, which
is exactly how electrostatic dust precipitators and photocopier toner work.
The energy. The energy of the configuration — the work needed to carry the charge in
from infinity — is
W = -\frac{q^2}{16\pi\varepsilon_0\,d}.
Notice the 16\pi, not 8\pi. This is
half of what two real charges +q and
-q a distance 2d apart would have,
-q^2/(8\pi\varepsilon_0 d). The factor of two is the subtle heart of the
images method — see the "Watch out!" box.
It feels like cheating that a single fictitious -q stands in for the whole
messy rearrangement of conduction electrons. The uniqueness theorem is what makes it honest. The real
conductor does just one thing that matters to the region above it: it holds its surface at
\varphi = 0. Anything — real charges, imaginary charges, a wish —
that produces a potential which solves Laplace's equation above the plane and is zero on the
plane must, by uniqueness, give the identical field there. The image charge is simply the cheapest
gadget that reproduces that one boundary value. It doesn't matter that the electrons and the image
bear no physical resemblance; they are indistinguishable from outside.
The same idea scales to prettier geometries. A charge outside a grounded sphere of radius
R at distance a from the centre is imaged by a
charge q' = -qR/a placed a distance R^2/a from
the centre — a smaller, off-centre mirror charge that again zeroes the potential on the sphere. Two
grounded planes meeting at a right angle need three image charges; parallel planes need an infinite
train of them. The mirror trick is a whole toolbox.
Two traps snare almost everyone the first time they meet image charges.
The image is fictitious, and only lives in the real region. The two-charge potential
is the correct answer only for z > 0, the region where the real
charge lives. Below the plane there is no field at all — a grounded conductor screens
it completely, so \varphi = 0 throughout z < 0.
If you blindly evaluate the two-charge formula at z < 0 you get a
perfectly definite number that is physically wrong. The image is a computational scaffold for
the real half-space, never a description of what is behind the mirror.
The energy is not the energy of two real charges. You might expect
W = -q^2/(8\pi\varepsilon_0 d), the interaction energy of
+q and -q a distance
2d apart. The true answer is half that,
-q^2/(16\pi\varepsilon_0 d). The reason: as you push the real charge in, the
image obediently moves too, but no work is done on the image — it isn't real, there is no
field energy stored in the fictitious lower half-space (where the true field is zero). Only the upper
half-space holds energy, so you collect only half the naive value. Compute image energies from the
work or the field integral over the real region, never by copying the two-real-charge formula.
The other great technique: separation of variables
Images are magic when a charge sits near a simple conductor, but many boundary-value problems have no
convenient mirror — a charge inside a rectangular box, the potential in a slot between biased plates,
a sphere with a patterned surface potential. For these the workhorse is
separation of variables. You look for solutions of Laplace's equation of the product
form
\varphi(x,y,z) = X(x)\,Y(y)\,Z(z),
which turns the one partial differential equation into three ordinary ones (each factor is a sine,
cosine, or exponential in Cartesian coordinates; Legendre polynomials and powers of
r in spherical coordinates). Because Laplace's equation is linear,
you may add such product solutions freely, and you tune the coefficients of the resulting
Fourier (or Legendre) series to match the boundary conditions term by term. Uniqueness
again guarantees that once the series fits the boundary, it is the answer. Images and separation of
variables are the two pillars of analytic electrostatics; between them they crack a remarkable share of
the classic problems.