LINEAR ALGEBRA AND VECTOR ANALYSIS

MATH 22A

Unit 14: Partial differential equations

Lecture
14.1. A partial differential equation is a rule which combines the rates of changes of
different variables. Our lives are affected by partial differential equations: the Maxwell
equations describe electric and magnetic fields E and B. Their motion leads to the
propagation of light. The Einstein field equations relate the metric tensor g with the
mass tensor T . The Schrödinger equation tells how quantum particles move. Laws
like the Navier-Stokes equations govern the motion of fluids and gases and especially
the currents in the ocean or the winds in the atmosphere. Partial differential equations
appear also in unexpected places like in finance, where for example, the Black-Scholes
equation relates the prices of options in dependence of time and stock prices.
14.2. If f (x, y) is a function of two variables, we can differentiate f with respect to
both x or y. We just write fx (x, y) for ∂x f (x, y). For example, for f (x, y) = x3 y + y 2 ,
we have fx (x, y) = 3x2 y and fy (x, y) = x3 + 2y. If we first differentiate with respect to
x and then with respect to y, we write fxy (x, y). If we differentiate twice with respect
to y, we write fyy (x, y). An equation for an unknown function f for which partial
derivatives with respect to at least two different variables appear is called a partial
differential equation PDE. If only the derivative with respect to one variable appears,
one speaks of an ordinary differential equation ODE. An example of a PDE is
fx2 + fy2 = fxx + fyy , an example of an ODE is f 00 = f 2 − f 0 . It is important to realize
that it is a function we are looking for, not a number. The ordinary differential equation
f 0 = 3f for example is solved by the functions f (t) = Ce3t . If we prescribe an initial
value like f (0) = 7, then there is a unique solution f (t) = 7e3t . The KdV partial
differential equation ft + 6f fx + fxxx = 0 is solved by (you guessed it) 2sech2 (x − 4t).
This is one of many solutions. In that case they are called solitons, nonlinear waves.
Korteweg-de Vries (KdV) is an icon in a mathematical field called integrable systems
which leads to insight in ongoing research like about rogue waves in the ocean.
14.3. We say f ∈ C 1 (R2 ) if both fx and fy are continuous functions of two variables
and f ∈ C 2 (R2 ) if all fxx , fyy , fxy and fyx are continuous functions. The next theorem is
called the Clairaut theorem. It deals with the partial differential equation fxy = fyx .
The proof demonstrates the proof by contradiction. We will look at this technique
a bit more in the proof seminar.

Theorem: Every f ∈ C 2 solves the Clairaut equation fxy = fyx .

 Linear Algebra and Vector Analysis

14.4. Proof. We use Fubini’s

R x0 +h R y0 +h theorem which will appear later in the double integral
lecture: integrate x0 ( y0 fxy (x, y) dy)dx by applying the fundamental theorem
R x +h
of calculus twice x00 fx (x, y0 + h) − fx (x, y0 ) dx = f (x0 + h, y0 + h) − f (x0 , y0 + h) −
R y +h R x +h
f (x0 +h, y0 )+f (x0 , y0 ). An analogous computation gives y00 ( x00 fyx (x, y) dx)dy =
f (x0 + h, y0 + h) − f (x0 , y0 + h) − f (x0 + h, y0 ) + f (x0 , y0 ). Fubini applied to g(x, y) =
R y +h R x +h R x +h R y +h
fxy (x, y) assures y00 ( x00 fyx (x, y) dx)dy = x00 ( y00 fyx (x, y) dy)dx so that
RR
f −fyx dydx = 0. Assume there is some (x0 , y0 ), where F (x0 , y0 ) = fxy (x0 , y0 )−
A xy
fyx (x0 , y0 ) = c > 0, then also for small h, Rthe
R function F is bigger than c/2 everywhere
on A = [x0 , x0 + h] × [y0 , y0 + h] so that A
F (x, y) dxdy ≥ area(A)c/2 = h2 c/2 > 0
contradicting that the integral is zero.
14.5. The statement is false for functions which are only C 1 . The standard counter
example is f (x, y) = 4xy(y 2 − x2 )/(x2 + y 2 ) which has for y 6= 0 the property that
fx (0, y) = 4y and for x 6= 0 has the property that fy (x, 0) = −4x. You can see the
comparison of f (x, y) = 2xy = r2 sin(2θ) and f (x, y) = 4xy(y 2 − x2 )/(x2 + y 2 ) =
r2 sin(4θ). The later function is not in C 2 . The values fxy and fyx , changes of slopes
of tangent lines, turn differently.

Figure 1. Clairaut holds for f (x, y) = 2xy which is in polar coordi-

nates r2 sin(2θ). It does not for the function f (x, y) = 4xy(y 2 −x2 )/(x2 +
y 2 ) which is in polar coordinates 2r2 sin(2θ) cos(2θ) = r2 sin(4θ).

Illustration
14.6. In many cases, one of the variables is time for which we use the letter t and
keep x as the space variable. The differential equation ft (t, x) = fx (t, x) is called
the transport equation. What are the solutions if f (0, x) = g(x)? Here is a cool
derivation: if Df = f 0 is the derivative, 1 we can build operators like (D+D2 +4D4 )f =
f 0 +f 00 +4f 0000 . The transport equation is now ft = Df . Now as you know from calculus,
the only solution of f 0 = af, f (0) = b is beat . If we boldly replace the number a with
with the operator D we get f 0 = Df and get its solution
eDt g(x) = (1 + Dt + D2 t2 /2! + · · · )g(x) = g(x) + g 0 (x)t + g 00 (x)t2 /2! + · · · .
By the Taylor formula, this is equal to g(x+t). You should actually remember Taylor
as g(x + t) = eDt g(x) . We have derived for g(x) = f (0, x) in C 1 (R2 ):
1We usually write df for derivative but D tells it is an operator. D also stands for Dirac.
 Theorem: ft = fx is solved by f (t, x) = g(x + t).
Proof. We can ignore the derivation and verify this very quickly: the function satisfies
f (0, x) = g(x) and ft (t, x) = fx (t, x). QED.
14.7. Another example of a partial differential equation is the wave equation ftt =
fxx . We can write this (∂t + D)(∂t − D)f = 0. One way to solve this is by looking at
(∂t − D)f = 0. This means transport ft = fx and f (t, x) = f (x + t). We can also have
(∂t +D)f = 0 which means ft = −fx leading to f (x−t). We see that every combination
af (x + t) + bf (x − t) with constants a, b is a solution. Fixing the constants a, b so that
f (x, 0) = g(x) and ft (x, 0) = h(x) gives the following d’Alembert solution. It
requires g, h ∈ C 2 (R).
g(x+t)+g(x−t) h(x+t)−h(x−t)
Theorem: ftt = fxx is solved by f (t, x) = 2
+ 2
.

14.8. Proof. Just verify directly that this indeed is a solution and that f (0, x) = g(x)
and ft (0, x) = h(x). Intuitively, if we throw a stone into a narrow water way, then the
waves move to both sides.
14.9. The partial differential equation ft = fxx is called the heat equation. Its
solution involves the normal distribution
2 2
√
N (m, s)(x) = e−(x−m) /(2s ) / 2πs2
in probability theory. The number m is the average and s is the standard deviation.
14.10. If the initial heat g(x) = f (0, x) at time t = 0 is continuous and zero outside a
bounded interval [a, b], then
Rb √
Theorem: ft = fxx is solved by f (t, x) = a g(m)N (m, 2t)(x) dm.
√
Proof. For every fixed m, the function N (m, 2t)(x) solves the heat equation.

f=PDF[ N o r m a l D i s t r i b u t i o n [m, Sqrt [ 2 t ] ] , x ] ; Simplify [D[ f , t ]==D[ f , { x , 2 } ] ]
 
Pn
Every Riemann sum approximation√g(x) = (1/n) k=1 g(mk ) of g defines a function
fn (t, x) = (1/n) nk=1 g(mk )N (mk , 2t)(x) which solves the heat
P
R ∞ equation. So does
f (t, x) = limn→∞ fn (t, x). To check f (0, x) = g(x) which need −∞ N (m, s)(x) dx = 1
R∞
and −∞ h(x)N (m, s)(x) dx → h(m) for any continuous h and s → 0, proven later.
14.11. For functions of three variables f (x, y, z) one can look at the partial differential
equation ∆f (x, y, z) = fxx + fyy + fzz = 0. It is called the Laplace equation and ∆ is
called the Laplace operator. The operator appears also in one of the most important
partial differential equations, the Schrödinger equation
~2
i~ft = Hf = − ∆f + V (x)f ,
2m
where ~ = h/(2π) is a scaled Planck constant and V (x) is the potential depending
on the position x and m is the mass. For i~ft = P f with P = −i~D, then the
solution f (x − t) is forward translation. The operator P is the momentum operator
in quantum mechanics. The Taylor formula tells that P generates translation.
 Linear Algebra and Vector Analysis

Homework

Problem 14.1: Verify that for any constant b, the function

f (x, t) = e−bt cos(x + t)
satisfies the driven transport equation
ft (x, t) = fx (x, t) − bf (x, t) .
This PDE is sometimes called the advection equation with damping b.

√
Problem 14.2: √ We have seen that f (t, x) = N (m, 2t) =
−(x−m)2 /(4t)
e / 4πt solves the heat equation ft = fxx . Verify more gen-
erally that √
2
e−(x−m) /(at) / aπt
solves the heat equation
ft = (a/4)fxx .

Problem 14.3: The Eiconal equation fx2 + fy2 = 1 can be rewritten as

||df || = 1, where df = ∇f = [fx , fy ]T is the gradient of f . (The gradient
is the transpose of the Jacobian matrix for the map f : R2 → R.) It is
an important equation in optics. Let f (x, y) be the distance to the circle
x2 + y 2 = 1. Show that it satisfies the eiconal equation.

Problem 14.4: The differential equation

ft = f − xfx − x2 fxx
is a version of the Black-Scholes equation. Here f (x, t) is the price of
a call option and x is the stock price and t is time. Find a function
f (x, t) solving it which depends both on x and t. Hint: look first for
solutions f (x, t) = g(t) or f (x, t) = h(x) and then for functions of the
form f (x, t) = g(t)h(x).

Problem 14.5: The partial differential equation

ft + f fx = fxx
is called Burgers equation and describes waves at the beach. In higher
dimensions, it leads to the Navier-Stokes equation which is used to de-
scribe the weather. Verify that the function
2
1 3/2 − x4t


xe
f (t, x) = qt 2
1 − x4t
t
e +1
is a solution of the Burgers equation. You better use technology.

Oliver Knill, [email protected], Math 22a, Harvard College, Fall 2018