oo
HARMONIC
FUNCTIONS
This chapter is devoted to the Laplace equation. We introduce two of its
important properties, the maximum principle and the rotational invariance.
Then we solve the equation in series form in rectangles, circles, and related
shapes. The case of a circle leads to the beautiful Poisson formula.
6.1 LAPLACE’S EQUATION
If a diffusion or wave process is stationary (independent of time), then u, = 0
and u,, = 0. Therefore, both the diffusion and the wave equations reduce to the
Laplace equation:
Uy. =0 in one dimension
V+ Vu=Au= uy, + Uy =0 in two dimensions
V+ Vu=Au= uy t uy + u,,=0 in three dimensions.
A solution of the Laplace equation is called a harmonic function.
In one dimension, we have simply u,,, = 0, so the only harmonic functions
in one dimension are u(x) = A + Bx. But this is so simple that it hardly gives us
a clue to what happens in higher dimensions.
The inhomogeneous version of Laplace’s equation
Au=f, a)
with fa given function, is called Poisson’s equation.
Besides stationary diffusions and waves, some other instances of Laplace’s
and Poisson’s equations include the following.
1466.1 LAPLACE’S EQUATION 147
Electrostatics. From Maxwell’s equations, one has curlE=0 and
divE = 4zp, where p is the charge density. The first equation implies E =
—grad @ for a scalar function ¢ (called the electric potential). Therefore,
Ad = div(grad d) = — divE = —4zp,
which is Poisson’s equation (with f= — 4p).
Steady fluid flow. Assume that the flow is irrotational (no eddies) so that
curly = 0, where v = v(x, y, 2) is the velocity at the position (x, y, z), as-
sumed independent of time. Assume that the fluid is incompressible (e.g.,
water) and that there are no sources or sinks. Then div v = 0. Hence v =
—grad ¢ for some ¢ (called the velocity potential) and Ad = — div v = 0,
which is Laplace’s equation.
Analytic functions of a complex variable. Write z= x + iy and
S(z) = u(z) + iv(z) = u(x + iy) + io(x + iy),
where wand vare real-valued functions. An analytic function is one that is
expressible as a power series in z. This means that the powers are not xy”
but 2" = (x + iy)". Thus
s@)= 5 ae
(a, complex constants). That is,
u(x + iy) + iv(x+ iy) = Yay (x + iy.
0
ie
Formal differentiation of this series shows that
ou _ ov ou __ ov
ax ay My ax
(see Exercise 1). These are the Cauchy- Riemann equations. If we differen-
tiate them, we find that
Dy = Uy 5
so that Au = 0. Similarly Av = 0, where Ais the two-dimensional laplacian.
Thus the real and imaginary parts of an analytic function are harmonic.
Brownian motion. Imagine brownian motion ina container D. Thismeans
that particles inside D move completely randomly until they hit the
boundary, when they stop. Divide the boundary arbitrarily into two pieces,
C, and C, (see Figure 1). Let u(x, y, z) be the probability that a particle
which begins at the point (x, y, z) stops at some point of C,. Then it can be
deduced that
Ug = Vy =
Au=0inD
u=lonC, u=0onC,.
Thus w is the solution of a Dirichlet problem.148 CHAPTER 6 HARMONIC FUNCTIONS
Figure 1
As we discussed in Section 1.4, the basic mathematical problem is to solve
Laplace’s or Poisson’s equation in a given domain D witha condition on bdy D:
Au=f inD
ou
u=h or ne or gn tua h on bdy D.
In one dimension the only connected domain is an interval {a = x = 5}. Wewill
see that what is interesting about the two- and three-dimensional cases is the
geometry.
MAXIMUM PRINCIPLE
We begin our analysis with the maximum principle, which is easier for La-
place’s equation than for the diffusion equation. By an open set we mean a set
that includes none of its boundary points (see Section A.1).
Maximum Principle. Let D bea connected bounded open set (in either two-
or three-dimensional space). Let either u(x, y) or u(x, y, z) be a harmonic
function in D that is continuous on D = D U (bdy D). Then the maximum and
the minimum values of u are attained on bdy D and nowhere inside (unless
u= constant).
In other words, a harmonic function is its biggest somewhere on the
boundary and its smallest somewhere else on the boundary.
To understand the maximum principle, let us use the vector shorthand
x = (x, y) in two dimensions or x = (x, y, z) in three dimensions. Also, the
radial coordinate is written as |x|= (x? + y?)!/2 or [x|= (x2 + yp? + 22)12,
The maximum principle asserts that there exist points x,y and x, on bdy D
such that
U(X) S U(X) S u(y) (2)
for all x € D (see Figure 2). Also, there are no points inside D with this property
(unless u = constant). There could be several such points on the boundary.6.1 LAPLACE’S EQUATION 149
*
Figure 2
The idea of the maximum principle is as follows, in two dimensions, say. At
a maximum point inside D, if there were one, we'd have u,, = 0 and uy, = 0.
(This is the second derivative test of calculus.) So u,, + uy = 0. At most maxi-
mum points, u,. <0 and Uyy <0. So we’d get a contradiction to Laplace’s
equation. However, since it is possible that u,,=0=u,, at a maximum
point, we have to work a little harder to get a proof.
Here we go. Let € > 0. Let v(x) = u(x) + €|x|2. Then, still in two dimen-
sions, say,
Av=Aut+eA(x?+y2)=0+4e>0 inD.
But Av=v,, + vy, <0 at an interior maximum point, by the second deriva-
tive test in calculus! Therefore, v(x) has no interior maximum in D.
Now v(x), being a continuous function, has to have a maximum some-
where in the closure D = D U bdy D. Say that the maximum of v(x) is attained
at Xp € bdy D. Then, for all x € D,
u(x) < v(x) = v(x) = u(Xp) + €|Xo|? = max ut €l?,
bay D
where /is the greatest distance from bdy D to the origin. Since this is true for any
€> 0, we have
u(x)Smaxu —forallxe D. (3)
bdy D
Now this maximum is attained at some point xy, € bdy D. So u(x) = u(x,4) for
all x € D, which is the desired conclusion.
The existence of a minimum point x,, is similarly demonstrated. (The
absence of points inside D will be proved by a different method in Section 6.3.)
a
UNIQUENESS OF THE DIRICHLET PROBLEM
To prove the uniqueness, suppose that
Au=f inD Av=f inD
u=h onbdyD v=h onbdyD.150 CHAPTER 6 HARMONIC FUNCTIONS
We want to show that u =v in D. So we simply subtract equations and let
w=u-— v. Then Aw=0 in Dand w= 0 on bdy D. By the maximum principle
0 = W(X) = W(X) = wW(Xyy) = 0 for all x € D.
Therefore, both the maximum and minimum of w(x) are zero. This means that
w=Oandu=v.
INVARIANCE IN TWO DIMENSIONS
The Laplace equation is invariant under all rigid motions. A rigid motion in the
plane consists of translations and rotations. A translation in the plane is a
transformation
x’=xt+a yv=ytsb.
Invariance under translations means simply that u,, + Uyy = Ux + Uyy,
A rotation in the plane through the angle a is given by
x’ =xcosatysina
; j (4)
y’ =-xsinat+ycosa.
By the chain rule we calculate
Ux, = Uys COS O — Uyr SID
Uy = Uys SID A + Uy COS
Ux = (Ux: COS O — Uy SiN O)y/ COS & — (Uy COS A — Uy Sin Q),, Sin a
Uyy = (Uy Sin a + Uy, COS @),/ Sin & + (Ux, SiN O& + Uy COS C)yr COS OL,
Adding, we have
Urge F Uyy = (Urges + Uyy) (Costar + sin2a) + Uyey + (0)
= Upigs + Uy.
This proves the invariance of the Laplace operator. In engineering the laplacian
Ais a model for isotropic physical situations, in which there is no preferred
direction.
The rotational invariance suggests that the two-dimensional laplacian
a2 a2
= 24"
2x2” ay?
should take a particularly simple form in polar coordinates. The transformation
x=rcos@ y=rsin@
has the jacobian matrix
ax ay
or dr} _( cos@ sin@
ax ay —rsin@ rcos@
00 (00
$=6.1 LAPLACE’S EQUATION 151
with the inverse matrix
or 36 —sin 6
St cos @ ———
pls Ox ox ae r
or 36 sing £2 @
oy oy, r
(Beware, however, that dr/ax # (8x/dr)—!.) So by the chain rule we have
oO 9 sind a
ae a ae
8 ing 8 4 £088 8
oy or r 3"
These operators are squared to give
[ogg 9 2 sind ap
axe [8 a 88
a sin 0cos@\ 02
= cos? 92 — 2 (SBP SOs’
ose or a ( r ) or a0
sin? @ 8? | 2sin @cos 0 oy sin? 6 9
v2 ag r 00 r or
a2 ~ 9 , cos@\ (a \?
= (snod+ : ) (3)
2 i 2
= sin? 925 +2 sin @cos@\ @
or?
r or 40
cos? @ 6? _2sinA@cos@ a | cos? a.
r2 ag? r 00 ror’
Adding these operators, we get (lo and behold!)
a2 a2 _ a? 10,1 &
2 Gxt ay? dr? rar r2 ag
(5)
It is also natural to look for special harmonic functions that themselves are
rotationally invariant. In two dimensions this means that we use polar coordi-
nates (r, @) and look for solutions depending only on r. Thus by (5)
i
OF Ue + Uyy = Upp + Uy
if u does not depend on @. This ordinary differential equation is easy to solve:
(ru,), = 0, ru, =, u=c,logr+cp.
The function log r will play a central role later.152 CHAPTER 6 HARMONIC FUNCTIONS
INVARIANCE IN THREE DIMENSIONS
The three-dimensional laplacian is invariant under all rigid motions in space.
To demonstrate its rotational invariance we repeat the preceding proof using
vector-matrix notation. Any rotation in three dimensions is given by
x’ = Bx,
where B is an orthogonal matrix (‘8B = B‘B = J). The laplacian is Au =
Shai ui = EPj—1 5,aij where the subscripts on w denote partial derivatives.
Therefore,
Au=5) (z busyby er => Sa ther
Kl Nii kl
=> Men
k
because the new coefficient matrix is
DY Puidyby = LPribu= (BBs = Fa-
a i
So in the primed coordinates Au takes the usual form
AU = Uggs + Uyy FH Uyegr
For the three-dimensional laplacian
2 a2 g2
~ ax” yh” az?
it is natural to use spherical coordinates (r, 0, @) (see Figure 3). We'll use the
notation
43
r= x24 y2 + 72 = vst $22
s=vx2+yi
x=scosé z=rcosé
y=ssing s=rsin 0.
Figure 36.1 LAPLACE'S EQUATION 153
(Watch out: In some calculus books the letters @ and 6 are switched.) The
calculation, which is a little tricky, is organized as follows. The chain of vari-
ables is (x, y, z) (8, 6, z) > (7, @, 8). By the two-dimensional Laplace calcu-
lation, we have both
1 1
Ugg + Uys = Ure + Ur + 5 Mog
and
a 1 1
Ug + Uy = igs Ts + Uae,
We add these two equations, and cancel w,,, to get
Ag = Ug, + Uyy + Use
ae 1 1
F Ure Fo Uy a Moa + Us + a Mag
In the last term we substitute s? = r? sin? @ and in the next-to-last term
ou or 06 ab
aoe
Us a5 — Mr as M095 + 95
=u Stuy 24 uy +0
r
This leaves us with
1
gu = ty +2 0, +S [vor (ot ue + 5 apt), (6)
which may also be written as
a 20 1 a. a 1 a2
=Stisct+aaa sta a
Agu a rar sind 00. : 00 + 1 sin? 6 62
Finally, let’s look for the special harmonic functions in three dimensions
which don’t change under rotations, that is, which depend only on r. By (7) they
satisfy the ODE
(7)
O= Ayu= uy +2 u,.
So (r?u,), = 0. It has the solutions r2u, = c,. That is, w= —c,r—! + cy. This
important harmonic function
i- (x2 + y2 $ 22)-172
is the analog of the special two-dimensional function log(x? + y2)!/2 found
before. Strictly speaking, neither function is finite at the origin. In electrostatics154 CHAPTER 6 HARMONIC FUNCTIONS
the function u(x) = r~! turns out to be the electrostatic potential when a unit
charge is placed at the origin. For further discussion, see Section 12.2.
EXERCISES
1. Show that a function which is a power series in the complex variable
x+ iy must satisfy the Cauchy-Riemann equations and therefore La-
place’s equation.
2. Find the solutions that depend only on r of the equation u,. + uy, +
Uzz = k2u, where k is a positive constant. (Hint: Substitute u = v/r.)
3. Find the solutions that depend only on r of the equation 1, + yy = k?u,
where k is a positive constant. (Hint: Look up Bessel’s differential equa-
tion in [MF] or in Section 10.5.)
4, Solve u, + uyy + uz, = 0 in the spherical shell O 0 for some n. The Y equation has
exponential solutions. As usual it is convenient to write them as
Y(y) =A cosh £,y + B sinh B,y.
So 0= Y’0)+ Y(O) = BB, +A. Without losing any information we
may pick B = —1, so that A = £,. Then
Y(y) = B, cosh B,y — sinh By. (4)
Therefore, the sum
u(x, Y) = A, sin B,x (By cosh £,y — sinh By) (5)
n=
is a harmonic function in D that satisfies all three homogeneous BCs. The
remaining BC is u(x, b) = g(x). It requires that
e(xy= > A, (B, cosh £,b — sinh £,5) - sin B,x
n=0
for0