2501-MTL603

ANANTA KUMAR MAJEE

1. Introduction
A partial differential equation (PDE in short) is an equation involving an unknown function
of two or more independent variables and its partial derivatives. PDEs occur naturally in
applications; they model the rate of change of a physical quantity with respect to both space
variables and time variables.
Definition 1.1. A multi-index α is an n-tuple of non-negative integers, say α = (α1 , α2 , . . . , αn ).
For a multi-index α, we define
i) |α| = ni=1 αi , α! = α1 !α2 ! · · · αn !.
P
ii) for any vector x = (x1 , x2 , · · · , xn ), we set xα = xα1 1 xα2 2 · · · xαnn .
iii) the differential operator
∂ |α|
Dα = .
∂xα11 ∂xα22 · · · ∂xαnn
∂f ∂2f
For α = (1, 0), Dα f = ∂x , and for α = (1, 1), Dα f = ∂x∂y where f is a function of x and y.
Definition 1.2. A PDE is an equation involving an unknown function of two or more variables
and certain of its partial derivatives.
• The order of the PDE is the highest order of derivatives involved.
• An expression of the form
F (~x, Du(~x), · · · , Dk u(~x)) = 0, ~x ∈ Ω ⊂ Rn (1.1)
k
is called a k-th order PDE where k ≥ 1 is an integer and F : Ω × R × Rn × · · · × Rn → R
is given and u : Ω → R is the unknown.
• The degree of a differential equation is the power of the highest derivative which occurs
in it, after the differential equation has been made free from radicals and fractions as far
as the derivatives are concerned.
Example 1.1. Examples of PDEs
i) ux (x, y) + uy (x, y) = 0 for (x, y) ∈ R2 .
ii) ut + uux = 0, t > 0, x ∈ R (Burger’s equation).
2
iii) ∆u = 0, where ∆u = ni=1 ∂∂xu2 (Laplace equation).
P
i
iv) ut − ∆u = 0, t > 0, ~x ∈ Rn (Heat equation).
v) utt − ∆u = 0 (Wave equation).
vi) ut + uux + uxxx = 0, t ≥ 0, x ∈ R (Korteweg-de Vries equation).
Definition 1.3. We say that u is a solution (classical solution) of the k-th order PDE (1.1), if
all partial derivatives involved exist and satisfies the PDE (1.1).
Remark 1.1. In general we look for solutions which satisfies certain boundary condition or
initial condition.
Definition 1.4. A k-th order PDE (1.1) is called
1
2 A. K. MAJEE

a) linear if it has the form |α|≤k aα (~x)Dα u = f (~x),

P

b) semi-linear if it is of the form |α|=k aα (~x)Dα u + a0 (~x, u, Du, . . . , Dk−1 u) = 0,

P

c) quasilinear if it is of the form

X
aα (~x, u, Du, . . . , Dk−1 u)Dα u + a0 (~x, u, Du, . . . , Dk−1 u) = 0,
|α|=k

d) fully nonlinear if it depends nonlinearly upon the highest order derivatives.

Example 1.2. Let us consider the various types of PDE:
a) Linear PDEs:
ux + uy = 0; xux + yuy = 0; ∆u = f (x); ut − ∆u = f (x).
b) Semi-linear PDEs:
ux + uy = u2 ; a(x, y)ux + b(x, y)uy = f (x, y, u).
c) Quasilinear PDEs:
a(x, y, u)ux + b(x, y, u)uy = c(x, y, u), ut + uux = 0,
∇u

div p = 0 (minimal surface equation).
1 + |∇u|2
d) Fully nonlinear PDE:
ux uy = u, u2x + u2y = 1 (Eikonal equation).
Definition 1.5. We say that a PDE is well-posed if
i) it has a solution
ii) solution is unique
iii) solution depends continuously on the given data, i.e., if the given data varies, then the
solution may not varies widely.
1.1. First order PDE & method of characteristic. For simplicity, we consider the PDEs
in two independent variables. We begin with a simple example, called transport equation.
Example 1.3 (Transport equation). Consider the problem of finding u(t, x) satisfying
ut + aux = 0, t > 0, x ∈ R,
(1.2)
u(0, x) = h(x) , x ∈ R,
where a ∈ R is a constant.
Case I: a = 0. Then the equation is ut = 0. Hence u(t, x) = h(x) ∀x ∈ R. So, to get a solution
at (t, x), we project this on to x-axis and take the initial value at this point, i.e., we are traveling
back along the lines parallel to t-axis to the initial curve to identify the solution.
Case II: a 6= 0. Note that the left hand side of (1.2) is the directional derivative of u(t, x) in
the direction (1, a). Find a curve in R2 such that it passes through the point (t, x) with slope
1 and a.Consider the function g : R → R2 as g(s) = (t + s, x + as). This line hits the plane
Γ := {t = 0} × R when s = −t at the point (0, x − at). Then F (s) = u(g(s)) satisfies
F 0 (s) = ut + aux = 0.
Therefore, F (·) is a constant function of s, and consequently for each point (t, x), u is constant
on the line through (t, x) with the direction (1, a). In particular, we have
u(t, x) = F (0) = F (−t) = u(g(−t)) = u(0, x − at) = h(x − at) .
Conversely, every u of the form u(t, x) = h(x − at) is a solution of (1.2) with initial values h
provided h is of class C 1 (R).
 PDES 3

Example 1.4 ( Non-homogeneous transport equation). Consider the non-homogeneous problem

ut + aux = f (t, x) t > 0, x ∈ R,

(1.3)
u(0, x) = h(x) , x ∈ R.
For fixed (t, x), define F (s) = u(t + s, x + as). Then, we have
F 0 (s) = ut + aux = f (t + s, x + as) .
Integrating, we have
Z −t Z t
F (0) − F (−t) = − f (t + s, x + as) ds = f (s, x + a(s − t)) ds.
0 0
On the other hand, F (0) − F (−t) = u(t, x) − u(0, x − at) = u(t, x) − h(x − at). Therefore
Z t
u(t, x) = h(x − at) + f (s, x + a(s − t)) ds .
0

Exercise 1.1. Find the solution of the PDE: ux + uy = u in R2 with initial data u(x, 0) = g(x)
for x ∈ R.
1.1.1. Method of characteristics: Consider the quasilinear equation in two independent vari-
ables x, y
aux + buy = c , (1.4)
where a, b and c are continuous functions in x, y and u. Let u(x, y) be a solution and z = u(x, y)
be the graph of u. Let z0 = u(x0 , y0 ). Then the normal to the surface

S := (x, y, z) : z = u(x, y)
at any point (x0 , y0 , z0 ) is N0 = (ux (x0 , y0 ), uy (x0 , y0 ), −1). Observe that the vector
V0 = (a(x0 , y0 , z0 ), b(x0 , y0 , z0 ), c(x0 , y0 , z0 ))
is perpendicular to the normalN0 , and hence V0 must lie on the tangent plane to the surface S
at (x0 , y0 , z0 ).
Aim: Find the surface z = u(x, y) knowing that the vector field V (x, y, z) = (a, b, c) lies on
the tangent plane of the surface S at the point (x, y, z). Such a surface is called the integral
surface. Thus, to find a solution of (1.4), we should try to find an integral surface.
Cauchy problem: Given a space curve Γ in R3 , find a function u(x, y) satisfying (1.4) such
that the level surface z = u(x, y) contains Γ. In other words, find u(x, y) satisfying
aux + buy = c in U ; u(x, y) = h(x, y) on Γ, (1.5)
where U is an open domain that contains the curve Γ.
Let the initial curve Γ is parameterized by (f (s), g(s), h(s)). For each fixed s, we construct
an integral surface S parameterized by s and t as S = {(x(s, t), y(s, t), z(s, t)) : t ≥ 0} so that at
t = 0, it coincides with the initial parameterization. Since (a, b, c) is perpendicular to the normal
to the surface, it is natural that it satisfies the system of equations with initial conditions:

dx
 dt = a(x(s, t), y(s, t), z(s, t)) , x(s, 0) = f (s)

dy
dt = b(x(s, t), y(s, t), z(s, t)) , x(s, 0) = g(s) (1.6)
 dz

dt = c(x(s, t), y(s, t), z(s, t)) , x(s, 0) = h(s)
We can solve this system of equations uniquely (via Picard’s theorem) for all small t under the
assumption that a, b and c are C 1 function. The curves so obtained are called characteristic
4 A. K. MAJEE

curves. To obtain the surface in the variables x, y, we need to find a function H such that
(s, t) = H(x, y). In this case, the unique solution of (1.4) is given by
u(x, y) = z(s, t) = z(H(x, y)).
In order to determine when we can do so, we make use of the Inverse Function Theorem.
Let G : U ⊂ Rn → Rn be a C 1 -function with G = (G1 , G2 , · · · , Gn ). We define the Jacobian of
G at the point ~x0 ∈ Rn as
 ∂G1
x0 ) ∂G x0 ) . . . ∂G

∂x1 (~ ∂x2 (~ ∂xn (~ x0 )
1 1

 ∂G2 (~x0 ) ∂G2 (~x0 ) . . . ∂G2 (~x0 ) 

 ∂x ∂x2 ∂xn
JG(~x0 ) = det  1 . .

.. .. .. ..
 . . . 
∂Gn ∂Gn ∂Gn
∂x1 (~
x0 ) ∂x2 (~
x0 ) ... ∂xn (~
x0 )

Theorem 1.1 (Inverse function theorem). Assume that G : U ⊂ Rn → Rn is a C 1 -function

and JG(~x0 ) 6= 0. Then there exist an open set V ⊂ U with ~x0 ∈ V and open set W ⊂ Rn
with ~z0 = G(~x0 ) ∈ W such that G : V → W is one-to-one and onto, and the inverse function
G−1 : W → V is C 1 .
Let G(s, t) = x(s, t), y(s, t) . Then G is C 1 and for any fixed s0 ∈ R,

∂
∂s x(s0 , 0)
∂
∂t x(s0 , 0) f 0 (s0 ) a(f (s0 ), g(s0 ), h(s0 ))
JG(s0 , 0) = = = bf 0 (s0 ) − ag 0 (s0 ) .
∂
∂s y(s0 , 0)
∂
∂t y(s0 , 0)
g 0 (s0 ) b(f (s0 ), g(s0 ), h(s0 ))
Hence, if JG(s0 , 0) 6= 0, by the inverse function theorem, there

exists an open set V containing
(s0 , 0) and open set W containing G(s0 , 0) = f (s0 ), g(s0 ) such that G : V → W is one-to-one
and onto, and the inverse function G−1 : W → V is C 1 . In other words, if bf 0 (s0 ) − ag 0 (s0 ) 6= 0,
then the Cauchy problem (1.5) has a solution in a nbd. of the initial curve Γ.
Definition 1.6 (Non-characteristic curve). A curve Γ is called non-characteristic if
bf 0 (s) − ag 0 (s) 6= 0.
Geometrically, this means that the tangent to Γ and the vector field (a, b, c) along Γ project to
vectors in the xy-plane are nowhere parallel.
In view of the above discussions, we arrive at the following theorem.
Theorem 1.2. Let Γ be a non-characteristic curve, and a, b, c are C 1 functions. Then the
Cauchy problem (1.5) has a solution in a nbd. of the initial curve Γ.
Example 1.5. Solve the initial value problem (IVP)
ux + 2uy = u2 x ∈ R, y > 0 ; u(x, 0) = h(x) , x ∈ R .
Solution: A parametrization of the initial curve Γ is {(s, 0, h(s)) : s ∈ R}. The characteristic
equations are
xt = 1, x(s, 0) = s; yt = 2, y(s, 0) = 0; zt = z 2 , z(s, 0) = h(s) .
Note that the Jacobian J 6= 0. Therefore, Γ is non-characteristic. Solving the characteristic
equations , we have
1 1
y = 2t, x = t + s, − = t − .
z h(s)
Inverting the variables, we get
h(s)
t = y/2, s = x − y/2, z = .
1 − th(s)
 PDES 5

The characteristic lines are y = 2x − 2s and the solution is

h(x − y/2)
u(x, y) =
1 − y2 h(x − y/2)
which is defined upto 1 − y2 h(x − y/2) 6= 0.
Example 1.6 (Characteristic problem). Consider the problem
ux + uy = 0 (x, y) ∈ R2 ; u = h on Γ .
Show that
i) Solution exists uniquely if Γ is not parallel to y = x.
ii) If Γ is parallel to y = x, the solution exists if and only if h is constant.
Solution: If Γ is parallel to y = x, the initial parametrization of Γ is {(s, s, h(s)) : s ∈ R} and
hence f 0 (s) = 1 = g 0 (s). Thus the Jacobian J = 0. Therefore, solution exists uniquely if Γ is
NOT parallel to y = x.
ii) Let us solve the problem with u(x, 0) = h̄(x). Then solving the characteristic equations, we
have
x = t + s , y = t , z = h̄(s) hence u(x, y) = h̄(x − y) .
Therefore, on y = x, u(x, x) = h̄(0). If u is a solution of the underlying problem, then it is
necessary and sufficient that h(x) = h̄(0) i.e., h is constant. In this case, there are infinitely
many solutions exist, a family of them is given by
u(x, y) = c + h̃(x − y) with h̃(0) = 0 .
Remark 1.2. The main difference between the semilinear and quasilinear equation is that the
first two equations in the system of characteristic ODEs decouples from the third equation in
the semilinear case, but it does not in the quasilinear case. Consequently, in the quasilinear
case, we may get projected characteristic curves crossing themselves. Let us see how that affects
the existence of solutions.
Example 1.7 (Burger’s equation). Consider the Burger’s equation
uy + uux = 0 x ∈ R, y > 0; u(x, 0) = h(x) . (1.7)
Here, the initial parametrization of the initial space curve is (s, 0, h(s)). The characteristic
equations are
xt = z, x(s, 0) = s; yt = 1, y(s, 0) = 0; zt = 0, z(s, 0) = h(s) .
Therefore, z = h(s), x = h(s)t + s and y = t. Hence the characteristic lines are x = h(s)y + s
and the speed of the curve is dx
dy = h(s). The solution can be defined implicitly as

u(x, y) = h(x − u(x, y)y) .

Let γ1 , γ2 be characteristic curve at s1 resp. at s2 with s1 < s2 speed h(s1 ) resp. h(s2 ). If they
intersect, then
s1 − s2
s1 + h(s1 )y = s2 + h(s2 )y =⇒ y = .
h(s2 ) − h(s1 )
Thus, if h is decreasing function, then the characteristic lines intersect at a point (x0 , y0 ) with
y0 > 0. This phenomena is called Gradient Catastrope or blow up. Since u is constant
along the characteristic, we have
u(x0 , y0 ) = u(s1 , 0) = h(s1 ), and u(x0 , y0 ) = u(s2 , 0) = h(s2 ).
But, by assumption h(s1 ) > h(s2 ). Therefore, we get a contradiction!
6 A. K. MAJEE

0
h (s)
Now from the solution u = h(x − uy), we see that ux = 1+yh 0
0 (s) . Hence if h (s) < 0, we find

that ux becomes infinite at the positive time y = − h01(s) . Thus, if h0 (s0 ) < 0, at any point s0 ,
then the solution does not exist globally. We can interpret the above as follows:
• If the initial velocity u(x, 0) of the fluid flow form a non-decreasing function of position,
then the fluid moves out in a smooth fashion.
• If the initial velocity is decreasing function, then the fluid flow undergo a shock that
correspond to collision of particles i.e., the integral surface folds itself.
Example 1.8. Consider the Burger’s equation as in Example 1.7 with initial condition h(x)
given by

1 , x < 0,

h(x) = 1 − x , x ∈ [0, 1),

0, x > 1.


In this case, the characteristic lines are z = h(s), x = h(s)t + s and y = t. So,

s + t , s < 0,

x(s, t) = s + t(1 − s) , s ∈ [0, 1],

s, s > 1.


For y < 1, the characteristic lines do not intersect. So, given a point (x, y) with y < 1, we can
draw the backward through characteristics

x − y , x < y < 1 ,

s = x−y
1−y , y ≤ x ≤ 1,

x, x > 1.


Therefore, the solution for y < 1 is given by


1 , x < y < 1 ,

u(x, y) = 1 − x−y
1−y , y ≤ x ≤ 1,

0, x > 1.


Propagation of discontinuity: There are situations where the initial condition has a dis-
continuity. In this case, we can not expect the solution to be C 1 . One such problem is called
Riemann problem. To understand the propagation of discontinuity along the characteristic
curve, let us consider the Burger’s equation with initial velocity h(x) given by
(
1 x < 0,
h(x) =
0, x > 0.
In this case, the characteristic curves are
(
y + s, s < 0,
x = h(s)y + s =
s, s > 0.
The characteristic lines intersect for all x > 0. Moreover, u(x, y) = 1 on the characteristic curve
x = y + s for s < 0 and u(x, y) = 0 for x > 0.
Let us consider another initial condition h1 :
(
0, x < 0
h1 (x) =
1, x > 0.
 PDES 7

Then, the characteristic curves are

(
s, s < 0,
x = h(s)y + s =
y + s, s > 0.
The characteristic lines do not intersect. One can ask the following question: how to define the
solution in this type of situations?. To answer this, one needs to define so called generalized
solutions or weak solutions.
1.2. General solution of quasilinear 1st order PDE:. In ODEs, an IVP is often solved
by finding a general solution that depends on an arbitrary constant and then using the initial
condition to evaluate the constant. For 1st order quasilinear PDE, a similar process may be
achieved by the method of Lagrange.
Definition 1.7 (General solution). F (φ, ψ) = 0, where φ = φ(x, y, z), ψ = ψ(x, y, z) and F is
an arbitrary smooth function, is called a general solution of f (x, y, z, p, q) = 0 if z, p and q as
determined by the relation F (φ, ψ) = 0 satisfies the PDE f (x, y, z, p, q) = 0.
Theorem 1.3. Suppose there exist two functions φ and ψ such that they are constant along the
characteristic equations of the quasilinear PDE aux + buy = c. Then F (φ, ψ) = 0 is a general
solution of the PDE, where F is such that Fφ2 + Fψ2 6= 0.
Proof. Let (x(t), y(t), z(t)) be the solution of the characteristic equation of the pde, and let
φ(x(t), y(t), x(t)) = c1 and ψ(x(t), y(t), x(t)) = c2 for some constant c1 and c2 . Then we have
φx dx + φy dy + φz dz = 0 , ψx dx + ψy dy + ψz dz = 0 .
dx dy dz
Since φ and φ are constant along a = b = c , we have
a −b c
= = . (1.8)
φy ψz − ψy φz φx ψz − φz ψx φx ψy − φy ψx
Since z = z(x, y), F is a function of x and y. Since F = 0, we see that Fx = 0 and Fy = 0 i.e.,
Fφ (φx + φz zx ) + Fψ (ψx + ψz zx ) = 0 ; Fφ (φy + φz zy ) + Fψ (ψy + ψz zy ) = 0 .
Since Fφ2 + Fψ2 6= 0, we have
φx + φz zx ψx + ψz zx
= 0 i.e., (φz ψy − φy ψz )zx + (φx ψz − φz ψx )zy = φy ψx − φx ψy
φy + φz zy ψy + ψz zy
Then using (1.8), we get that azx + bzy = c. Thus, the general solution is given by
F (φ, ψ) = 0.

Remark 1.3. Since F should satisfy only condition, + Fφ2 Fψ2 6= 0, one may choose F is the
form
F (φ, ψ) = φ + g(ψ)
for arbitrary smooth function g.
Example 1.9. Find a general solution of uux + yuy = x.
Solution: The characteristic equation in the non-parametric form can be written as
dx dy dz
= = . (1.9)
z y x
From (1.9), we have xdx − zdz = 0 i.e., d(x2 − z 2 ) = 0. Therefore, take φ(x, y, z) = x2 − z 2 .
Then φ(x, y, z) is constant along (1.9). Now by using (1.9), we see that
xdy − ydx = ydz − ydx = ydz − zdy
8 A. K. MAJEE

y
=⇒ (x + z)dy − yd(x + z) = 0 =⇒ d( ) = 0.
x+z
y
Therefore, we take ψ(x, y, z) = x+z . Then ψ(x, y, z) is constant along the characteristic equa-
tions. Thus, the general solution is F (φ, ψ) = φ + g(ψ) = 0, i.e.,
y
u2 = x2 + g( )
x+u
for arbitrary smooth function g.
Remark 1.4. For nonlinear equations, the term general solution need not mean that all solutions
are of this form. This phenomenon should be familiar from ODEs. For example, the general
√ 2
solution of ux + uy = u is given by u(x, y) = (x+f (x−y))
4 for arbitrary smooth function f . But
the trivial solution u ≡ 0 is not covered by the general solution.
1.3. Nonlinear equation: A general nonlinear 1st order PDE in x, y takes of the form
F (x, y, u, ux , uy ) = 0.
Let p = ux and q = uy . Suppose F has a quasilinear form
F ≡ a(x, y, z)p + b(x, y, z)q − c(x, y, z) = 0.
Then the characteristic equations are
dx dy dz
Fp = a = , Fq = b = , pFp + qFq = ap + bq = c = .
dt dt dt
Taking this as motivation, we write three equations, for general nonlinear 1st order PDE
dx dy dz
= Fp , = Fq , = pFp + qFq .
dt dt dt
We need equations satisfies by p and q as well. Observe that
dp d dx dy
= ux = uxx + uxy = Fp px + Fq qx .
dt dt dt dt
We do not want px and qx . To do so, we differentiate F = 0, with respect to x and get the
following equation:
Fx + Fz zx + Fp px + Fq qx = 0 =⇒ Fp px + Fq qx = −Fx − Fz zx = −Fx − Fz p .
Thus,
dp
= −Fx − Fz p.
dt
Similarly, we have
dq
= −Fy − Fz q.
dt
Thus, the five equations
dx


 = Fp
dt


dy




 = Fq
 dt


dz
= pFp + qFq
 dt
dp


= −Fx − Fz p




 dt
 dq

 = −Fy − Fz q

dt
form the characteristic strip. The initial parametrization of the given initial curve Γ gives the
initial conditions for x, y and z. To solve the above system of ODEs, we need to find initial
 PDES 9

values for p and q. Observe that, on Γ, h(s) = u(f (s), g(s)). Thus, if p0 (s) and q0 (s) be the
initial values for p and q resp. then it should satisfy the strip condition
p0 (s)f 0 (s) + q0 (s)g 0 (s) = h0 (s) , (1.10)
and the admissible condition, i.e., F (f (s), g(s), h(s), p0 (s), q0 (s)) = 0 on the initial curve. So,
p0 and q0 are such that
(
p0 (s)f 0 (s) + q0 (s)g 0 (s) = h0 (s) ,
(1.11)
F (f (s), g(s), h(s), p0 (s), q0 (s)) = 0 .
Observe that, in order to construct the integral surface S, we are interested in the support of
the strip, namely the curve (x(t), y(t), z(t)), but to find it, we need to find the functions p(t)
and q(t). Suppose Γ is non-characteristic, i.e.,
f 0 (s)Fq (f (s), g(s), h(s), p0 (s), q0 (s)) − g 0 (s)Fp (f (s), g(s), h(s), p0 (s), q0 (s)) 6= 0 .
Then, by using Inverse function theorem, we will be able to invert the function G(s, t) =
(x(s, t), y(s, t)) and get the unique solution of the given nonlinear PDE in a nbd. of Γ. Thus,
we arrive at the following theorem.
Theorem 1.4. If Γ is non-characteristic for nonlinear problem F (x, y, z, p, q) = 0 and functions
p0 (s) and q0 (s) exist and satisfy the strip and admissible conditions (1.11), then there is an
integral surface S containing Γ (which is unique for the choice of p0 and q0 ).
Example 1.10. Solve the IVP
ux uy = u , x, y ∈ R; u(0, y) = y 2 .
Solution: The problem can be written as F (x, y, u, ux , uy ) = 0 where F (x, y, z, p, q) = pq − z.
Parametrization of initial curve is {(0, s, s2 ) : s ∈ R}. Initial functions p0 and q0 must satisfy
the conditions (1.11), i.e., q0 (s) = 2s and p0 (s) = 2s . The characteristic strip satisfy
dx dy dz
= q, x(s, 0) = 0; = p, y(s, 0) = s; = 2pq = 2z, z(s, 0) = s2 ,
dt dt dt
dp s dq
= p, p(s, 0) = ; = q, q(s, 0) = 2s .
dt 2 dt
Solving the above equations, we have
s s
q(s, t) = 2set , p(s, t) = et , z(s, t) = s2 e2t , x(s, t) = 2s(et − 1) , y(s, t) = (1 + et ).
2 2
t 1 x
Notice that se = 2 ( 2 + 2y). Hence the solution is given by
x
u(x, y) = z = (set )2 = (y + )2 .
4
Example 1.11. Solve the IVP
3
uy = u3x , x, y ∈ R; u(x, 0) = 2x 2 , x ∈ R .
3
Solution: Here F (x, y, z, p, q) = p3 − q. Parametrization of the initial curve is {(s, 0, 2s 2 ) :
3
s ∈ R}. Therefore, f (s) = s, g(s) = 0 and h(s) = 2s 2 . From the strip condition, we have
1 3
p0 (s) = 3s 2 . Again from the admissible condition, we get q0 (s) = 27s 2 . The characteristic strip
are
dx dy dz 3
= 3p2 , x(s, 0) = s; = −1, x(s, 0) = 0; = 3p3 − q , z(s, 0) = 2s 2 ,
dt dt dt
dp 1 dq 3
= 0, p(s, 0) = 3s 2 ; = 0 , q(s, 0) = 27s 2 .
dt dt
10 A. K. MAJEE

Solving the above ODEs, we have

3


q(s, t) = 27s 2 ,
1


p(s, t) = 3s 2 ,


y(s, t) = −t ,

x(s, t) = 27st + s ,




 3 3
z(s, t) = 2s 2 + 54s 2 t .

x
Note that from x(s, t) and y(s, t), we have s = 1−27y . Thus, the solution u(x, y) is given by
3
3 x2 3 1
u(x, y) = z = 2s (1 − 27y) = 2
2
3 (1 − 27y) = 2x 2 (1 − 27y)− 2 .
(1 − 27y) 2

1.4. Complete integral and general solutions: We have considered general solution for
quasilinear problems. Do such general solutions exist for fully nonlinear equations? The answer
is yes but the process is more complicated than the quasilinear case. Let us first consider so
called complete integrals. Let A ⊂ R2 be an open set which is the parameter set. For any
C 2 -function u, a = (a1 , a2 ) and x = (x, y), we denote
 
2 ua1 uxa1 uya1
(Da u, Dxa u) = .
ua2 uxa2 uya2
Definition 1.8 (Complete integral). A C 2 - function u(x, a) is said to be a complete integral in
U × A of the equation F (x, y, u, ux , uy ) = 0 in U if
i) u(x, a) solves the PDE F (x, y, u, ux , uy ) = 0,
2 u) is equal to 2.
ii) rank of (Da u, Dxa
Example 1.12. Find a complete integral of ux uy = u.
Solution: From the given equation, we have F (x, y, z, p, q) = pq − z. The characteristic equa-
tions are
dx dy dz dp dq
= q, =p. = 2z = p, = q.
dt dt dt dt dt
From last two equations, we have p = c1 et and p = c2 et . Thus from third equation, we have
z = c1 c2 e2t + c3 . From the first equation, we have x = c2 et + a and from the second equation, we
have y = c1 et + b. Thus, (x − a)y − b = c1 c2 e2t , and hence u(x, y, a, b) = z = (x − a)(y − b) + c3 .
So, u(x, y, a, b) will be a solution if c3 = 0. Then we get
u(x, y, a, b) := (x − a)(y − b).
It is easy to check that
 
2 b−y 0 −1
(Da u, Dxa u) = ,
−x + a −1 0
whose rank is 2. Therefore, u(x, y, a, b) = (x − a)(y − b) is a complete integral.
Remark 1.5. In general, complete integral is not unique. Moreover, all the solutions cannot
covered from the complete integral.
Next we study how to build more complicated solutions for nonlinear 1st order PDEs. We
construct these new solutions as envelopes of complete integrals.
Definition 1.9 (Envelope of a family). Suppose u(x, y, a1 , a2 ) be a C 1 function on U × A where
U is an open subset of R2 and A be the parameters set. If the equations
∂u
(x, y, a1 , a2 ) = 0 (x, y) ∈ U , (a1 , a2 ) ∈ A (i = 1, 2)
∂ai
 PDES 11

can be solved for the parameters and has solution a1 = φ(x, y), a2 = ψ(x, y) , i.e.,
∂u
(x, y, φ(x, y), ψ(x, y)) = 0 (x, y) ∈ U,
∂ai
then we call the function v(x, y) = u(x, y, φ(x, y), ψ(x, y)) the envelope of the functions {u(·, a)}a∈A .
Theorem 1.5. Suppose for each (a1 , a2 ) ∈ A, u = u(·, a1 , a2 ) solves the 1st order nonlinear
PDE F (x, y, u, ux , uy ) = 0. Assume that the envelope v defined as above exists and is a C 1 -
function. Then v(x, y) solves F (x, y, u, ux , uy ) = 0.
Proof. v(x, y) is a envelope of the family {u(·, a)}a∈A such that v(x, y) = u(x, y, φ(x, y), ψ(x, y)).
In view of the assumption, we see that
F (x, y, u(x, y, φ(x, y), ψ(x, y)), ux (x, y, φ(x, y), ψ(x, y)), uy (x, y, φ(x, y), ψ(x, y))) = 0 .
Suppose v is C 1 . Then
∂u ∂u
vx (x, y) = ux (x, y, φ(x, y), ψ(x, y)) + (x, y, φ(x, y), ψ(x, y))φx (x, y) + (x, y, φ(x, y), ψ(x, y))ψx (x, y)
∂a1 ∂a2
∂u
= ux (x, y, φ(x, y), ψ(x, y)) , as (x, y, φ(x, y), ψ(x, y)) = 0 .
∂ai
Similarly, vy (x, y) = uy (x, y, φ(x, y), ψ(x, y)). Thus, F (x, y, v(x, y), vx (x, y), vy (x, y)) = 0. This
completes the proof. 
Definition 1.10 (Singular solution). The solution v described in Theorem 1.5 is called singular
solution of the nonlinear 1st order PDE F (x, y, u, ux , uy ) = 0.
Example 1.13. Find the singular solution of u2x + u2y = 1 + 2u.
Solution: To find singular solution, we first need to find complete integral of the given PDE.
Here F (x, y, z, p, q) = p2 + q 2 − 1 − 2z. The characteristic equations are
dx dy dp dq
= 2p , = 2q , = 2p , = 2q .
dt dt dt dt
Solving these, we have p = c1 e2t , q = c2 e2t . Hence pq = a. Since p2 + q 2 − 1 − 2z = 0, we have
r r
1 + 2z 1 + 2z
p = ±a 2
, q=± .
1+a 1 + a2
Now from the strip condition
r r r
dz dx dy 1 + 2z dx 1 + 2z dy 1 + 2z  dx dy 
=p +q = ±a ± = ± a + .
dt dt dt 1 + a2 dt 1 + a2 dt 1 + a2 dt dt
√
Integrating, we have 1 + 2z = ± √ax+y 1+a2
+ b, and hence
1  ax + y 2 1
u(x, y, a, b) := √ +b − .
2 1 + a2 2
 2
One can check that rank of (Da u, Dxa 2 u) is 2. Hence u(x, y, a, b) = 1 √ax+y + b − 12 is a
2 1+a2
complete integral. Now ua = 0 and ub = 0 gives √ax+y + b = 0. Thus, v(x, y) = − 21 is a singular
1+a2
solution.
To generate more solutions from the complete integrals, we vary the above construction.
Choose an open set à ⊂ R and a C 1 - function h : à → R so that the graph (ã, h(ã)) lies with
in A ⊂ R2 .
Definition 1.11 (General integral). The general integral (depending on h) is the envelope
ṽ(x, y) of the functions {u(·, ã)}ã∈Ã provided this envelope exists and is C 1 , where u(x, y, ã) =
u(x, y, ã, h(ã)).
12 A. K. MAJEE

Example 1.14. Find a general integral of ux uy = u.

Solution: We have shown that a complete integral of the above PDE is given by
u(x, y, a, b) = (x − a)(y − b).
x+y
Let h : R → R be h(a) = a. Then u(x, y, a) = (x − a)(y − a), and hence ua = 0 gives a = 2 .
 2
Therefore, the general integral is v(x, y) = x − x+y x+y
x−y


2 y − 2 = − 2 .

Example 1.15. Find a general integral of ut + u2x = 0.

Solution: Given problem is a Hamilton-Jacobi equation with H(p) = p2 . Hence a complete
integral is given by
u(x, t, a, b) := ax − ta2 + b.
Let h : R → R be a function such that h(a) = a2 . Then u(x, y, a) = ax − ta2 + a2 , and hence
x x2
ua = 0 gives a = 2(t−1) . Therefore, the general integral is v(x, y) = 4(t−1) .
1.5. General Method of Characteristics for First-Order Equations. Consider the first-
order, nonlinear equation in n independent variables:
F (~x, u(~x), Du(~x)) = 0, ~x ∈ Rn ; u = φ on Γ, (1.12)
where Γ ⊂ Rn is (n−1) dimensional manifold. Note that an m-dimensional manifold is a surface
which can be represented
locally as the graph of a function.
Let { γ1 (~r), . . . , γn (~r) : ~r ∈ Rn−1 } be the parametrization of Γ. Let z(t) = u(~x(t)) and
pi (t) = uxi (~x(t)). Then the problem can be reformulated as F (~x, z, p~) = 0. We define 2n + 1
characteristic equations by
n
dxi dz X dpi
= Fpi , = pi Fpi , = −Fxi − Fz pi (1 ≤ i ≤ n) .
dt dt dt
i=1
The initial conditions are given by
xi (~r, 0) = γi (~r), z(~r, 0) = φ(~r), pi (~r, 0) = ψi (~r) (1 ≤ i ≤ n),
where the functions ψi are determined by solving the following equations:
n
X ∂γ1
φri (~r) = ψj (~r) (~r)
∂ri
j=1
F (γ1 (~r), . . . , γn (~r), φ(~r), ψ1 (r), . . . , ψn (~r)) = 0 .
In order to get the solution, we need to convert (~r, s) 7→ ~x(~r, s) near the manifold Γ. Let
G(~r, s) = ~x(~r, s). Then G is C 1 and for fixed ~r ∈ Rn−1 , we have

 ∂x1 ∂x1 ∂x1 

∂r1 (~r, 0) ∂r2 (~
r, 0) ... ∂t (~
r, 0)
 ∂x2 (~r, 0) ∂x2 ∂x2
 ∂r1 ∂r2 (~
r, 0) ... ∂t (~
r, 0) 
JG(~r, 0) = det 

.. .. .. .. 
 . . . . 
∂xn ∂xn ∂xn
∂r1 r , 0)
(~ ∂r2 r , 0) . . .
(~ ∂t (~r, 0)
 ∂γ ∂γ1

1
∂r1 (~r) ∂r2 (~
r) ... Fp1 ((γ1 (~r), . . . , γn (~r), φ(~r), ψ1 (r), . . . , ψn (~r))
∂γ2 ∂γ2
(~r) ∂r2 (~
r) ... Fp2 ((γ1 (~r), . . . , γn (~r), φ(~r), ψ1 (r), . . . , ψn (~r)) 
 
 ∂r1
= det  .. .. .. .. .
. . . .
 
 
∂γn ∂γn
∂r1 (~
r) ∂r2 (~
r) ... Fpn ((γ1 (~r), . . . , γn (~r), φ(~r), ψ1 (r), . . . , ψn (~r))
By inverse function theorem, if JG(~r, 0) 6= 0, then the Cauchy problem has a solution in a nbd.
of the manifold Γ.
 PDES 13

Remark 1.6. The manifold Γ is called non-characteristic, if JG(~r, 0) 6= 0. Note that if Γ =

{(x1 , x2 , . . . , xn ) : xn = 0} and the PDE is quasilinear PDE i.e.,
n
X
ai (~x, u)uxi + a0 (~x, u) = 0
i=1

then the non-characteristic condition of Γ is given by the condition

an ((x1 , x2 , . . . , xn−1 , 0), φ(x1 , x2 , . . . , xn−1 )) 6= 0.
Suppose u is a C 1 solution of the 1st order quasilinear PDE
n
X
ai (~x, u)uxi + a0 (~x, u) = 0
i=1
u(x1 , x2 , . . . , xn−1 , 0) = g(x1 , x2 , . . . , , xn−1 ) .

Can we compute ∇u on the plane {xn = 0} in terms of the given quantity? Note that since
u(x1 , x2 , . . . , xn−1 , 0) = g(x1 , x2 , . . . , , xn−1 ), we have
uxi (x1 , x2 , . . . , xn−1 , 0) = gxi (x1 , x2 , . . . , , xn−1 ), 1 ≤ i ≤ (n − 1).
To find uxn , we use the given PDE. Denote (x1 , x2 , . . . , xn ) := ~x = (x0 , xn ). We have
"n−1 #
X
an ((x0 , 0), g(x0 ))uxn ((x0 , 0)) = − ai ((x0 , 0), g(x0 ))gxi (x0 ) + a0 ((x0 , 0), g(x0 ))
i=1
"n−1 #
0 1 X
0 0 0 0 0
=⇒ uxn ((x , 0)) = − ai ((x , 0), g(x ))gxi (x ) + a0 ((x , 0), g(x ))
an ((x0 , 0), g(x0 ))
i=1

provided an ((x0 , 0), g(x0 )) 6= 0 i.e., the plane {xn = 0} is non-characteristic.

1.5.1. m-th order quasilinear initial value problem on flat boundary. Consider the
problem
X
aα (~x, u, . . . , Dm−1 u)Dα u + b(~x, u, . . . , Dm−1 u) = 0 ,
|α|=m

∂u 0 ∂ m−1 u
u(x0 , 0) = g0 (x0 ), (x , 0) = g1 (x0 ), . . . , m−1 (x0 , 0) = gm−1 (x0 ) .
∂xn ∂xn
Can we compute u and all its derivatives of order upto m for a C m solution of the above problem
on the plane {xn = 0}? Let α = (α1 , α2 , . . . , αn ) be a multi-index.
Case-1: |α| < m. If αn = 0, then Dα u = Dα g0 . If αn 6= 0, then 1 ≤ αn ≤ m − 1 and hence
∂ α1 . . . ∂ αn−1 ∂ αn u ∂ α1 +...+αn−1
Dα u = αn−1 = αn−1 gαn .
∂xα1 1 · · · ∂xn−1 ∂xαnn ∂xα1 1 · · · ∂xn−1
Case=II: |α| = m. If αn = 0, then Dα u = Dα g0 . If αn 6= 0, then 1 ≤ αn ≤ m. If αn ≤ m − 1,
m
then Dα u can be obtained by differentiating gαn . S0, it remains to compute only ∂∂xmu . From
n
∂mu
the PDE, we see that ∂xm can be computed if
n

aα0 ((x0 , 0), g0 (x0 ), . . . , Dm−1 u(x0 , 0)) 6= 0

i.e., the plane {xn = 0} is non-characteristic, where α0 = (0, 0, . . . , m).
14 A. K. MAJEE

1.6. Weak solution and scalar conservation laws: We will study the IVP
ut − (f (u))x = 0, t > 0, x ∈ R ,
(1.13)
u(x, 0) = h(x) , x ∈ R,
where f : R → R is a continuous function. Method of characteristic demonstrates that there
does not in general exist a smooth solution of (1.13), existing for all times t > 0. We therefore
look for some sort of weak solution or generalized solution.
Let us explain why the problem is called conservation laws. Suppose the problem has a
solution. Then integrating the PDE over the interval [a, b], we obtain
d b
Z Z b Z b
u(x, t)dx = ut (x, t)dx = − (f (u))x dx = f (u(a, t)) − f (u(b, t))
dt a a a
= [inflow at a] − [outflow at b] .
In other words, the quantity u is neither created nor destroyed; the total amount of u contained
inside any given interval [a, b] can change only due to the flux of u across the two end points.
Hence it is called conservation laws.
Let v : R × [0, ∞) → R be a smooth function with compact support. Then multiplying the
equation (1.13) by v and then integrating by parts formula, we get
Z ∞Z ∞ Z ∞Z ∞
0= ut v dx dt + (f (u))x v dx dt
0 −∞ 0 −∞
Z ∞Z ∞ Z ∞Z ∞ Z ∞
=− uvt dx dt − f (u)vx dx dt − h(x)v(x, 0) dx.
0 −∞ 0 −∞ −∞
Here v is called test function. Note that in the last expression, we need less regularity of the
function u (no derivative of u involved).
Definition 1.12 (Weak solution). Any bounded measurable function u(x, t) is called a weak
solution of (1.13) if the following holds:
Z ∞Z ∞ Z ∞
h(x)v(x, 0) dx = 0 ∀ v ∈ Cc∞ (R × [0, ∞)) .


uvt + f (u)vx dx dt + (1.14)
0 −∞ −∞

Remark 1.7. The followings hold:

i) All classical solutions are weak solutions but not all weak solutions are classical solutions.
In particular, if u ∈ C 1 (R × [0, ∞)) and is a weak solution i.e., satisfies (1.14), then u is
a classical solution of (1.13).
ii) Weak solution need not even be continuous.
Let u(x, t) be a weak solution of (1.13) such that it has a discontinuity across the curve x(t),
and u is smooth on either side of the curve. We claim that such a curve cannot be arbitrary.
Theorem 1.6 (Rankine-Hugoniot Condition). Let u(x, t) be a weak solution of (1.13) such that
it has a discontinuity across the curve x(t), and u is smooth on either side of the curve x(t).
Then u(x, t) must satisfies the Rankine-Hugoniot (R-H) condition
f (u− ) − f (u+ )
x0 (t) = ,
u− − u+
where u− (t, x) is the limit of u approaching (x, t) from the left of x(t) and u+ (x, t) is the limit
of u approaching (x, t) from the right of the curve x(t).
Proof. Let
Dl = {(x, t) : 0 < t < ∞, −∞ < x < x(t)} and Dr = {(x, t) : 0 < t < ∞, x(t) < x < ∞}.
 PDES 15

Let v ∈ Cc∞ (R × [0, ∞)). If u(x, 0) = 0, by divergence theorem, we gave

ZZ ZZ Z
u− η2 + f (u− )η1 v ds .


[uvt + f (u)vx ] dx dt = − [ut + (f (u))x ]v dx dt +
Dl Dl x=x(t)

where η = (η1 , η2 ) is the outward unit normal to the curve x(t) pointing from Dl to Dr . Similarly,
ZZ ZZ Z
u+ η2 + f (u+ )η1 v ds .


[uvt + f (u)vx ] dx dt = − [ut + (f (u))x ]v dx dt −
Dr Dr x=x(t)

Since u is a weak solution of (1.13) and ut + (f (u))x = 0 in Dl and Dr , we have from the above
two integral equations
Z Z
− −
u+ η2 + f (u+ )η1 v ds, ∀ v ∈ Cc∞ (R × [0, ∞)).



u η2 + f (u )η1 v ds =
x=x(t) x=x(t)

Thus,
f (u− ) − f (u+ ) η1 = u+ − u− η2



along x(t).
Now on the curve x = x(t) whose parametrization is (x(t), t), we take (η1 , η2 ) = 1
1+(x0 (t))2
(1, −x0 (t)),
and hence
f (u− ) − f (u+ )
x0 (t) = .
u− − u+
This completes the proof. 

Notation:
[[u]] = u− − u+ : jump in u across the curve x = x(t),
[[f (u)]] = f (u− ) − f (u+ ) : jump in f (u),
σ = x0 (t) = speed of the curve x = x(t)
Then the R-H condition reads as
[[f (u)]] = σ[[u]].
In fact in view of Remark 1.7, we have the following theorem.
Theorem 1.7. Let u : R × [0, ∞) → R be a piecewise C 1 function. Then u is a weak solution
of the problem (1.13) if and only if
a) u is a classical solution of (1.13) in the region where u is C 1 .
b) The R-H condition holds along each discontinuity line inside the domain.
Example 1.16 (Shock wave). Consider the Burger’s equation ut + uux = 0 in R × (0, ∞) with
initial condition g given by

1, if x ≤ 0,

g(x) = 1 − x, if 0 ≤ x ≤ 1,

0, if x > 1 .


We have seen that for t < 1, the classical solution (via method of characteristic) exists and given
by the formula

1, if x ≤ t,

u(x, t) = 1−x 1−t , if t ≤ x ≤ 1,

0, if x > 1 .

16 A. K. MAJEE

We need do define u as weak solution for t ≥ 1. Observe that the initial data for x < 0 wants
u = 1 , while the initial data for x > 1 wants u = 0 for t ≥ 1. Thus, we define
(
1, x < x(t)
u(x, t) =
0, x > x(t),
where x = x(t) is the curve of discontinuity. In addition, we want our curve x = x(t) to contain
the point (x, t) = (1, 1) (point of intersection of characteristic curves). By using R-H condition,
we have x0 (t) = 21 , and hence x = x(t) = 1+t 2 . Thus, weak solution of the given IVP is
(
1+t
1, x< 2 ,
u(x, t) = 1+t
0, x> 2 .

Example 1.17 (Weak solution for discontinuous initial data). Consider the IVP ut + uux = 0
with initial data
(
1, if x < 0,
h(x) =
0, if x > 0 .
Then for all x > 0, the characteristic intersects. We wand to define a weak solution of the
problem. Observe that, the initial data for x < 0 wants u = 1 and the initial data for x > 0
wants u = 0 for t > 0. Let x = x(t) be a curve of discontinuity such that it contains the point
(0, 0). Then from R-H condition, we have x0 (t) = 21 with x(0) = 0. Thus, x(t) = 2t . Thus, weak
solution of the underlying problem is given by
(
1, x < 2t ,
u(x, t) =
0, x > 2t .
Example 1.18 (Rarefaction waves and nonphysical shocks). Consider the Burger’s equation
ut + uux = 0 with initial condition
(
0, if x < 0,
h(x) =
1, if x > 0 .
In this case, the wave on the right moves faster than the wave on the left moves slower. So there
is no shock. We have
(
0, x < 0,
u(x, t) =
1, x > t.
There is no information in the interval [0, t]. One can then define continuous solution (called
Rarefaction wave solution) as

0,
 x ≤ 0,
u1 (x, t) = xt , 0 ≤ x ≤ t,

1, x > t.


Even one can introduce shock and define shock wave solution as
(
0, x < 2t ,
u2 (x, t) =
1, x > 2t .
In the second case, R-H condition holds as the speed of the discontinuity curve x(t) = 2t is 12 ,
and f (uul )−f (ur )
l −ur
= 12 = σ. This example shows the non-uniqueness of weak solutions. Moreover,
there exists infinitely many weak solution for the underlying problem: for λ ∈ (0, 1), define
 PDES 17


0,
 x < λ 2t ,
uλ (x, t) = λ, λ 2t ≤ x < (1 + λ) 2t ,
x ≥ (1 + λ) 2t .

1,


Each uλ is a weak solution to the Cauchy problem, because it satisfies the equation a.e and R-H
conditions holds along the two lines of discontinuity x1 (t) = λ 2t and x2 (t) = (1 + λ) 2t . To define
physically relevant solution, one needs to define so called entropy solution. For the Burger’s
equation, entropy condition reads as
f 0 (ul ) > σ > f 0 (ur ).
Note that for the solution u2 , ul = 0, ur = 1, σ = 21 , f 0 (ul ) = 0 and f 0 (ur ) = 1. Thus, u2 is
NOT a physically relevant solution.
Example 1.19 (Formation of secondary shock). Consider the Burger’s equation uy + uux = 0
with initial condition u(x, 0) = h(x), where
(
1, 0 < x < 1,
h(x) =
0, otherwise .
In this case, characteristic equations are
dy dx dz
= 1, y(s, 0) = 0; = z, x(s, 0) = s; = 0, z(s, 0) = h(s).
dt dt dt
Solving these, we have
x(s, t) = h(s)t + s, y = t, z = h(s).
If s < 0, then x = s and u = 0. If 0 < s < 1, then h(s) = 1, and hence x = y + s and u = 1. If
s > 1, then again x = s and u = 0. Therefore, we get rarefaction wave between x = 0 and x = y
and a shock is formed by the intersection of the lines x = y + s for 0 < s < 1 and x = 1. Let
x(y) be the curve of discontinuity so that it contains the point (1, 0). In this case, ul = 1 and
ur = 0. Hence from R-H condition, we have x0 (t) = 21 . Since x(0) = 1, we have x(y) = y2 + 1.
So our weak solution takes on the values


 0, x < 0,
 x , 0 < x < y,

u(x, y) = y


 1, y < x < 1 + y2 ,
0, x > 1 + y2


for y ≤ 2. Notice that at y = 2, the rarefaction wave hits the shock curve x = 1 + y2 . Therefore,
we need jump across the shock to satisfies the R-H condition. In this case, ul = xy and ur = 0.
x
Hence speed of the curve of discontinuity σ = 2y . Thus, from R-H condition, we obtain
dx x 1
= =⇒ log(x) = log(y) + C .
dy 2y 2
1 √
Since the curve contains the point (2, 2), we see that C = 2 log(2) and hence x(y) = 2y. Thus,
for y ≥ 2, weak solution takes on values

0,
 x < 0,
√
u(x, y) = xy , 0 < x < 2y,
 √
0, x > 2y .

18 A. K. MAJEE

2. Second order PDEs in two independent variables

A general second order quasilinear PDE in two independent variables x, y is of the form
auxx + 2bux,y + cuyy = d , (2.1)
where a, b, c and d are functions of x, y, u, ux and uy .
Cauchy Problem: Find u(x, y) satisfying (2.1) with given (compatible) values of u, ux and uy
on a given curve γ in the xy-plane. If γ has a parametrization (f (s), g(s)), then we prescribe,
on γ, the Cauchy data
u = h(s) , ux = φ(s) , uy = ψ(s) . (2.2)
In general, Cauchy data along γ is
∂u
u = h, = h1 ,
γ ∂η
γ

where η denotes a unit normal vector along γ, and ∂u

∂η = ∇u · η.
Note that, the values of any function v(x, y) and its first partial derivatives vx and vy along
the curve γ are connected by the compatibility condition (Strip condition)
dv
= vx f 0 (s) + vy g 0 (s)
ds
which follows by differentiating v(f (s), g(s)) with respect to s. Thus we have
h0 (s) = φ(s)f 0 (s) + ψ(s)g 0 (s) .
Compatibility condition also hold for the higher partial derivatives of any function on γ. Thus,
taking v = ux and v = uy , we find that
dux
= uxx f 0 (s) + uxy g 0 (s) ,
ds
duy
= uxy f 0 (s) + uyy g 0 (s) .
ds
Thus, if u(x, y) is a solution of (2.1) and (2.2), then
a uxx + 2b uxy + c uyy = d
f 0 uxx + g 0 uxy + 0 uyy = φ0
0 uxx + f 0 uxy + g 0 uyy = ψ 0 .
This determines uxx , uxy and uyy uniquely if
f 0 g0 0
2 2
∆ = 0 f 0 g 0 6= 0 i.e., ag 0 − 2bf 0 g 0 + cf 0 6= 0 .
a 2b c
Definition 2.1. The initial curve γ is called a characteristic curve ( with respect to the
6 0 along γ.
differential equation and data) if ∆ = 0. It is called non-characteristic if ∆ =
Along a non-characteristic curve, the Cauchy data uniquely determine the second derivatives
of u on γ. Moreover, we can obtain higher derivatives of u along γ. The Cauchy problem with
Cauchy data prescribed on a characteristic curve generally has no solution.
It is useful to express the characteristic condition in more geometrical and algebraic terms.
The principle symbol σ(ξ), associated to principle part of (2.1), defined by
σ(ξ) = σ(x, y; ξ) = a(x, y)ξ12 + b(x, y)ξ1 ξ2 + c(x, y)ξ22 , ξ = (ξ1 , ξ2 ).
 PDES 19

0 dy 0
Since γ has a parametrization (x = f (s), y = g(s)), we see that dxds = f (s) and ds = g (s), and
0 0
hence the vector ξ = (g , −f ) is normal to γ. Therefore, the curve γ is characteristic at (x, y)
if and only if the principal symbol vanishes on its normal vector ξ.
For further investigation, let us remove the parameter s by writing the characteristic condition
ady 2 − 2bdxdy + cdx2 = 0 .
dy
We can solve the above equation for dx in the form
√
dy b ± b2 − ac
= . (2.3)
dx a
Note that (2.3) is an ordinary differential equation for γ provided a, b and c are known functions
of x and y, i.e., the equation (2.1) is principally linear.
Definition 2.2. We say that the quasilinear PDE (2.1) is called
i) Elliptic if ac − b2 > 0 (there is no real characteristic curve).
ii) Hyperbolic if ac − b2 < 0. In this case, there are two families of characteristic curve.
iii) Parabolic if ac = b2 . In this case, only one characteristic curve is possible.
Remark 2.1. For nonlinear case, type (Elliptic, Parabolic, Hyperbolic) is not determined by
the differential equation but can depend on the individual solution. The ”type ” may change
with the point of the plane.
Example 2.1. Consider the equation uxx − uy = 0. Here, a = 1, b = 0 and c = 0. Hence
ac = b2 , and therefore, the equation is everywhere parabolic. The characteristic curve y = c
dy
found by dx = 0. This is one dimensional heat equation.
Example 2.2. Consider the PDE uxx − uyy = 0. Here, b = 0, a = 1, and c = −1, so b2 > ac.
dy
Thus, the equation is of hyperbolic type in the xy-plane. The characteristic equation is dx = ±1.
Hence the characteristic curves are given by y = ±x + c. The above equation is known as one
dimensional wave equation.
Example 2.3. Consider now the 2nd order PDE uxx + uyy = 0. Here, a = 1 = c and b = 0.
Equation is everywhere elliptic. There is no real characteristic. This equation is called Laplace
equation.
Example 2.4 (Tricomi equation). Consider the linear PDE uyy − yuxx = 0. Here b2 − ac = y.
a) The equation is of hyperbolic type in the upper half plane (i.e., y > 0). The characteristic
dy
equation is given by dx = ± √1y , and hence the equation of characteristic curves are
3
3x ± 2y 2 = c.
b) It is parabolic on the x-axis. The characteristic curve y = c.
c) It is of elliptic type in the lower half plane. There is no real characteristic.
Remark 2.2. Consider a general 2nd order PDE in two independent variables x, y as
F (x, y, u, ux , uy , uxx , uxy , uyy ) = 0.
Let
∂F 1 ∂F ∂F
a= , b= , c= .
∂uxx 2 ∂ux,y ∂uyy
Then the PDE F (x, y, u, ux , uy , uxx , uxy , uyy ) = 0 is hyperbolic, parabolic, elliptic if ac − b2 <
0 , ac − b2 > 0 , ac − b2 = 0 respectively.
Example 2.5. Consider the Monge-Ampère equation uxx uyy − u2xy = f (x). Here, a = uyy ,
b = −uxy and c = uxx . Thus,
i) equation is elliptic for a solution u exactly when f (x) > 0.
20 A. K. MAJEE

ii) equation is hyperbolic for a solution u exactly when f (x) < 0.

iii) equation is parabolic for a solution u exactly when f (x) = 0.
2.1. Canonical form of 2nd order principally linear PDE. Consider a 2nd order princi-
pally linear PDE
a(x, y)uxx + 2b(x, y)uxy + c(x, y)uyy = d(x, y, ux , uy ) . (2.4)
We use a change of coordinates from (x, y) to (ξ, η) so that equation (2.4) may transformed
into a equation so that its principal part takes the form of wave, heat or Laplace equation. The
Canonical forms may sometimes be used to find the general solution.
Lemma 2.1. Under a non-degenerate change of coordinates (x, y) 7→ (ξ, η), the equation (2.4)
to
A(ξ, η)uξξ + 2B(ξ, η)uξη + C(ξ, η)uηη = D̃(ξ, η, u, uξ , uη ) (2.5)
where, the coefficient functions A, B and C are given by
A(ξ, η) = aξx2 + 2bξx ξy + cξy2 , B(ξ, η) = aξx ηx + b(ξx ηy + ξy ηx ) + cξy ηy ,
C(ξ, η) = aηx2 + 2bηx ηy + cηy2 .
In fact the following equality holds:

2
B 2 − AC = ξx ηy − ξy ηx (b2 − ac) .


In view of the above lemma, if J = ξx ηy − ξy ηx , which is nonzero for non-degenerate change
of coordinates, we see that the type of the equation does not change.
Theorem 2.2 (For hyperbolic equation). Let the equation (2.4) by hyperbolic in a region Ω of
the xy-plane. Let (x0 , y0 ) ∈ Ω. Then there exists a change of coordinates (x, y) 7→ (ξ, η) in a
nbd. of (x0 , y0 ) such that (2.4) reduces to a 2nd order hyperbolic PDE of the form
uξη = D̃(ξ, η, u, uξ , uη ) . (2.6)
√
b2 −ac
ξ is constant along the characteristic curve determined by dy
the ODE dx = b(x,y)−
a(x,y) , and η is
√
2
b −ac
constant along the characteristic curve determined by the dy
ODE dx = b(x,y)+
a(x,y) .

Remark 2.3. Taking further change of variables x̃ = ξ + η and ỹ = ξ − η, the equation (2.8)
transformed into the PDE
uỹỹ − ux̃x̃ = D̃(x̃, ỹ, u, ux̃ , uỹ ).
Example 2.6. Find the canonical form of the PDE: x2 uxx − 2xyux,y − 3y 2 uyy + uy = 0.
Solution: In this case, a = x2 , b = −xy and c = −3y 2 . Thus, b2 − ac = 4 x2 y 2 . Therefore,
the equation is hyperbolic at every point (x, y) such that xy 6= 0. At the point on the coordinate
axes , the equation is of parabolic type. Let us consider the case x > 0, y > 0. Then equation is
of hyperbolic type there. The equation of characteristic curves dxdy
= −y±2y
x . Thus, the solutions
−1 −1
are x y = c and x y = c. Define ξ(x, y) = x y and η = η(x, y) = x3 y. Then we have
3

−16x2 y 2 uξη + 5x−1 uξ + x3 uη = 0 .

1
The above equation can be written in the variable ξ and η completely. We see that x = ( ηξ ) 4 and
1
y = (ξ 3 η) 4 . Substituting the values of x and y, we have
5 1 1 1
uξη − 1 uξ − uη = 0 .
16 (ξ 3 η 5 ) 4 16 (ξ 7 η) 14
This is required canonical form.
 PDES 21

Example 2.7. Find the general solution of the 2nd order PDE xuxx + 2x2 uxy = ux − 1.
Solution: Here a = x, b = x2 and c = 0. So, b2 − ac = x4 > 0 for x 6= 0. hence the equation
is hyperbolic provided x 6= 0. The characteristic curves are found by solving
√ (
dy x2 ± x4 2x
= =
dx x 0.

Hence we have y = x2 + c and y = c. Therefore, ξ(x, y) = x2 − y and η(x, y) = y. Then the

−3
equation reduces to 4x3 uξη = −1 and hence uξη = − 41 (ξ + η) 2 . This is desired canonical form.
−1
Now integrating with respect to η, we have uξ = 12 (ξ + η) 2 + f (ξ). Integrating again with respect
to ξ, we obtain
1
u = (ξ + η) 2 + F (ξ) + G(η).
Inverting to the variables x and y, we obtain our general solution as
u(x, y) = x + F (x2 − y) + G(y) ,
where F and G are arbitrary functions.
Theorem 2.3 (For parabolic equation). Let the equation (2.4) be parabolic in a region Ω of the
xy-plane. Let (x0 , y0 ) ∈ Ω. Then there exists a change of coordinates (x, y) 7→ (ξ, η) in a nbd.
of (x0 , y0 ) such that (2.4) reduces to a 2nd order hyperbolic PDE of the form
uηη = D̃(ξ, η, u, uξ , uη ) . (2.7)
dy b(x,y)
ξ is constant along the characteristic curve determined by the ODE dx = a(x,y) , and η is chosen
such a way that (ξ, η) defines a new coordinate system near (x0 , y0 ).
Example 2.8. Find the canonical form of the PDE
x2 uxx − 2xyuxy + y 2 uyy = 0.
Solution: In this case, a = x2 , b = −xy , c = y 2 , and hence the PDE if of parabolic type at
every point (x, y) ∈ R2 . Note that at (0, 0), the equation deduces to 0 = 0, and thus, we can
determine canonical form in any domain NOT containing the origin. In order to find the new
dy
coordinate system (ξ, η), we need to solve the ODE dx = − xy to find ξ and then choose η so that
(ξ, η) represents a coordinate system. Note here that ξ(x, y) = xy. Take η(x, y)=y so that the
Jacobian J = −x 6= 0. For this coordinate system, we have
ux = yuξ + uη ; uxx = y 2 uξξ + 2yuξη + uηη ; uy = xuξ ; uxy = xyuξξ + xuξη + uξ ; uyy = x2 uξξ .
Thus, the PDE reduces to
ξ
x2 uηη − 2xyuξ = 0 , i.e, uηη = 2 uξ .
η2
Theorem 2.4 (For elliptic equation). Let the equation (2.4) be elliptic in a region Ω of the
xy-plane. Let (x0 , y0 ) ∈ Ω. Then there exists a change of coordinates (x, y) 7→ (ξ, η) in a nbd.
of (x0 , y0 ) such that (2.4) reduces to a 2nd order hyperbolic PDE of the form
uηη + uξξ = D̃(ξ, η, u, uξ , uη ) . (2.8)
dy
In order to find the new coordinate (ξ, η), one needs to solve the characteristic ODE dx =
√
b(x,y)− b2 −ac
a(x,y) in the complex plane. Let Φ is constant along the characteristic. Then (ξ, η) is
given by ξ(x, y) = Re Φ(x, y) and η(x, y) = Im Φ(x, y).
22 A. K. MAJEE

Example 2.9. Find the canonical form of the PDE uxx + x2 uyy = 0.
Solution: Observe that the equation is of elliptic type at every point except on the y-axis. Let
dy 2
us solve the characteristic equation dx = ±ix in the complex plane. Note that Φ = y + i x2 is
2
constant along the characteristic. Set ξ(x, y) = Re Φ(x, y) = y and η(x, y) = Im Φ(x, y) = x2 .
For this coordinate system, we have
uxx = uη + x2 uηη , uyy = uξξ .
Therefore, required canonical form is
1
uξξ + uηη + uη = 0 .
2η
Here ξ = const. lines represents a family of straight lines parallel to x- axis and η = const. lines
represents family of parabolas.
Consider a second-order PDE in n space dimensions:
n n
X ∂2u X ∂u
Lu = aij (~x) + bi (~x) + c(~x)u = 0 . (2.9)
∂xi ∂xj ∂xi
i,j=1 i=1

~
Let aij = aji . The principle symbol of the second-order PDE (2.9) is a quadratic form in ξ;
~ := ξ~T A (~x)ξ,
Lp (~x, iξ) ~ A = ((−aij )).

Definition 2.3. Equation (2.9) is called

i) Elliptic if all the eigenvalues of A have the same sign,
ii) Parabolic if A is singular,
iii) Hyperbolic if all but one of the eigenvalues of A have same sign and one has the opposite
sign.
2.2. First order system: In ODEs, we have seen that 2nd order ODE can be reduced to 1st
order system. For PDEs, we may expect the same. Let us consider the PDE uxx − uyy = 0. We
know that it is of hyperbolic PDE. Take ~u = (u1 , u2 ), where u1 = ux and u2 = uy . Then the
PDE can be re-written as
 
0 −1
~uy + ~ux = ~0.
−1 0
Thus, we need to study the system of 1st order PDEs. A general system of first order PDEs in
R2 takes the form
~
A(x, y)~ux + B(x, y)~uy = C(x, y, ~u) (2.10)
~ ~u ∈ RN .
where A, B ∈ RN ×N , and C,
Question: What is the characteristic for this system? We define a characteristic to be a curve
γ in R2 for which the Cauchy data ~u does NOT determine the derivatives ~ux and ~uy along γ.
γ
Suppose γ is the graph of a function φ, so (s, φ(s)) parametrizes γ and ~u(s, φ(s)) = f~(s) is
the Cauchy data. Suppose ~u is a solution of (2.10). Then we have
~;
A~ux + B~uy = C ~ux + φ0 ~uy = f~0 .
Multiply the second equation by A and then subtract it from the first equation, we have (B −
φ0 A)~uy = ~c − Af~0 . Thus, we shall not be able to solve for ~uy if
dy dy
det(B − A) = 0 , where φ0 = . (2.11)
dx dx
 PDES 23

The equation (2.11) is the characteristic equation for (2.10). Geometrically, the curve γ is
characteristic for (2.10) at (x, y) if and only if the principal symbol matrix σ(x, y; ξ) = A(x, y)ξ1 +
B(x, y)ξ2 , where ξ = (ξ1 , ξ2 ) is singular on the vector ξ that is normal to γ.
Let us consider an initial value problem, where γ is a curve given by y = 0. In this case,
normal vector is (0, ξ2 ). Therefore, γ is non-characteristic if detB(x, 0) 6= 0 for all x ∈ R. By
continuity, detB(x, y) 6= 0 for sufficiently small y. Hence the equation (2.10) may assume in the
simpler form
~ near y = 0 .
~u + Ã(x, y)~u = C̃ (2.12)
y x

Thus, for (2.12), condition (2.11) becomes simply det( dx

dy I − Ã) = 0. Thus, to be a characteristic,
dx
the curve γ parametrized by (x(t), t) should satisfy dt = λ(x, t), where λ(x, t) is an eigen values
for Ã(x, t).
Question: What does hyperbolicity mean for (2.12)?
Definition 2.4. We say that the problem (2.12) is
i) Strictly hyperbolic if the N × N matrix à has N distinct real eigen values λ1 , λ2 , . . . , λN .
ii) Hyperbolic if the N × N matrix à has all real eigen values and a basis of eigen vectors.
Let Γ denote the matrix having the eigen vectors µ ~ N as column vectors. Then Γ−1
~ 1, . . . , µ
−1
exists and Γ ÃΓ = Λ, where Λ is the diagonal matrix with diagonal entries λ1 , . . . , λN . Take
~v = Γ−1 ~u. Then the equation (2.12) reduces to
~ y, ~v ) = Γ−1 C(x,˜y, ~u) −~Γy~v − ÃΓx~v .
~ y, ~v ) , where d(x,


~vy + Λ~vx = d(x, (2.13)
If ~u(x, 0) = f~(x) for (2.12), then ~v (x, 0) = Γ−1 f~(x). Sometimes, equation (2.13) me be decouples,
and then one can solve the PDE by solving the 1st-order PDE component-wise via method of
characteristic.
24 A. K. MAJEE

3. Cauchy-Kovalevski Theorem and Holmgren Uniqueness theorem

One may ask the following natural question: is there exist smooth solution of a given PDE
if the Cauchy data is smooth. The answer is no. But the partial answer to this question is
settled by the Cauchy-Kovalevski theorem— which asserts the local existence of a power series
solution of a general quasilinear PDE with given analytical data. We begin with the following
multinomial theorem.
Theorem 3.1 (Multinomial theorem). For any x = (x1 , x2 , · · · , xn ) ∈ Rn and any positive
integer k, one has

k X k!
x1 + x2 + · · · + xn = xα . (3.1)
α!
|α|=k

Proof. We prove it by induction. Let n = 2. Then, by the binomial theorem, we have

k
k!
xj1 xk−j
X
k
(x1 + x2 ) = 2 . (3.2)
j! (k − j)!
j=0

Let α = (α1 , α2 ) with α1 = j and α2 = k − j. Then |α| = k and xα1 1 xα2 2 = xα . Hence, from
(3.2), we see that
X k! X k!
(x1 + x2 )k = xα1 1 xα2 2 = xα .
α1 ! α2 ! α!
|α|=k |α|=k

Hence the result (3.1) holds for n = 2. Suppose the result is true for n − 1. If we show that
the result is true for n, then by the mathematical induction, it will be true for any n ∈ N. Let
x̃ = (x1 , x2 , · · · , xn−1 ). We have, by induction hypothesis,

k
k X k!
x1 + x2 + · · · + xn = (x1 + x2 + · · · xn−1 ) + xn = (x1 + x2 + · · · xn−1 )i xjn
i! j!
i+j=k
X k! X i! β j
= x̃ xn
i! j! β!
i+j=k |β|=i

for β = β1 , β2 , · · · , βn−1 ). Let α = (β1 , β2 , · · · , βn−1 , j). Then α! = β! j! and hence, we get

k X k!
x1 + x2 + · · · + xn = xα , as x = (x̃, xn ) .
α!
|α|=k


For any two smooth functions f, g : Rn → R, we have the following product rule for higher
order partial derivatives. This is Known as Leibniz formula.
X α!
Dα (f g) = (Dβ f )(Dγ g) .
β! γ!
β+γ=α
α
P
We will consider infinite multiple
P series of the form
P α cα x , where α varies over all multi-indices
and cα ∈ R. We say that α cα converges if α |cα | converges.
Example 3.1. Consider the following standard examples.
P α
i) α x is convergent if |xi | < 1 for all i. Indeed, one has
X X 1
xα = xα1 1 xα2 2 · · · xαnn = Qn .
α α i=1 (1 − xi )
 PDES 25

P |α| ! α
ii) α α! x is convergent for |x1 | + · · · + |xn | < 1. Moreover, by multinomial theorem,
X |α| ! ∞ X ∞
X k! α X
k 1
xα = x = x1 + x2 + · · · + xn = Pn .
α
α! α! 1 − i=1 xi
k=0 |α|=k k=0
P α! α−β
iii) The series β≤α (α−β)! x is convergent for |xi | < 1 for all i. Moreover,
X α! β!
xα−β = Qn 1+βi
.
(α − β)! i=1 (1 − xi )
β≤α

Lemma 3.2. If the power series α cα (x − x0 )α converges for some x̃ with |x̃i − (x0 )i | = Ri ,
P
n
then the power series convergesP uniformlyα in {x∞∈ R : |xni − (x0 )i | < Ri }. Moreover, the
function denoted by f (x) = α cα (x − x0 ) is C in {x ∈ R : |xi − (x0 )i | < Ri }.
Proof. Since α cα (x̃−x0 )α = α cα ni=1 (x̃i −(x0 )i )αi is convergent, we have α |cα | ni=1 Riαi <
P P Q P Q
+∞. If |xi − (x0 )i | < Ri , then
X X n
Y
|cα (x − x0 )α | ≤ |cα | Riαi < +∞ .
α α i=1

Hence the series α cα (x − x0 )α converges uniformly in {x ∈ Rn : |xi − (x0 )i | < Ri }.

P

To prove that f ∈ C ∞ in D := {x ∈ Rn : |xi − (x0 )i | < Ri }, we show that α cα Dβ xα

P
n
converges uniformly inP {x ∈ R α!: |xi − (x0 )α−β
i | < ri } with 0 < ri < RP
i for any
Qβ. We want to
show that the series α≥β cα (α−β)! (x − x0 ) is convergent. Since α |cα | ni=1 Riαi < +∞,
there exists a constant c > 0 such that
n n
Ri−αi .
Y Y
|cα | Riαi ≤ c =⇒ |cα | ≤ c
i=1 i=1

Hence, we have
n n
X α! X Y −α α! Y α −β
cα (x − x0 )α−β ≤ c Ri i ri i i
(α − β)! (α − β)!
α≥β α≥β i=1 i=1
n n 
α! ri αi −βi
Ri−βi
Y X Y
= c < +∞ ,
(α − β)! Ri
i=1 α≥β i=1

since Rrii < 1. Thus, β xα converges uniformly on {x ∈ Rn : |xi − (x0 )i | ≤ ri }. This

P
α cα D
completes the proof. 
Definition 3.1. Let f be a real-valued function defined on the open set Ω ⊂ Rn . We say that
f is real analytic at x0 ∈ Ω, if there exists a nbd. of x0 in which f can be expressed as a Taylor
series
X
f (x) = cα (x − x0 )α .
α

We say that f is real analytic in Ω if f is real analytic at every point x0 ∈ Ω.

Remark 3.1. If f is real analytic at x0 , then f is C ∞ in the nbd. of x0 . Moreover, the constant
cα is given by
Dα f (x0 )
cα = .
α!
26 A. K. MAJEE

Remark 3.2. In general, C ∞ - function may not be real analytic. For example, consider the
function
( 1
e− x2 , x > 0
f (x) =
0, x ≥ 0 .
One can check that it is a smooth function but not real analytic at x = 0.
Lemma 3.3. Let Ω be an open and connect set and f, g : Ω → R be two real analytic function
in Ω. Suppose there exists x0 ∈ Ω such that Dα f (x0 ) = Dα g(x0 ) for all α. Then f = g in Ω.
Proof. Let Ω1 = {x ∈ Ω : Dα f (x) = Dα g(x), ∀ α}. Then Ω1 6= ∅ as x0 ∈ Ω. Let x̃ ∈ Ω1 . Since
f is real analytic at x̃, we have, in the nbd. of x̃,
X Dα f (x̃) X Dα g(x̃)
f (x) = (x − x̃)α = (x − x̃)α = g(x) .
α
α! α
α!
Thus, f = g, and hence Dα f = Dα g in a nbd. of x̃. In other word, the nbd. of x̃ contained in Ω1
showing that Ω1 is open. On the other hand, Ω1 is the intersection of sets which are relatively
closed in Ω and hence Ω1 is closed set. Since Ω is connected, we conclude that Ω1 = Ω. This
completes the proof. 
Theorem 3.4. Let Ω ⊂ Rn be open and f ∈ C ∞ (Ω). Then f is real analytic in Ω if and only
if for every compact set K in Ω, there exist M > 0, r > 0 such that
|Dα f (x)| ≤ M |α|! r−|α| , ∀ x ∈ K, ∀ α . (3.3)

Remark 3.3. By using the fact that |α|! ≤ α! n|α| , and α! ≤ |α|! it is easy to see the equivalence
of (3.3) and the inequality
|Dα f (x)| ≤ M α! r−|α| for different M > 0 and r > 0.
Proof. Suppose f is real analytic in Ω . Fix x0 ∈ Ω. Since f is real analytic at x0 , there exists
R > 0 such that
X Dα f (x0 )
f (x) = (x − x0 )α , x ∈ {y : |yi − (x0 )i | ≤ R} .
α
α!
This implies that there exists Cx0 > 0 such that
|Dα f (x0 )| |α|
R ≤ Cx0 < +∞ .
α!
Let β be any multi index. Then, we have, for x ∈ {y : |yi − (x0 )i | ≤ R}
X Dα f (x0 ) α!
Dβ f (x) = (x − x0 )α−β
α! (α − β)!
α≥β
X |Dα f (x0 )| α!
=⇒ |Dβ f (x)| ≤ |(x − x0 )α−β |
α! (α − β)!
α≥β
X α!
≤ Cx0 R−|α| |(x − x0 )α−β |
(α − β)!
α≥β
R
Suppose x ∈ {y : |yi − (x0 )i | ≤ 2 }.Then, we get
X α! 1 X α! 1 1
|Dβ f (x)| ≤ Cx0 R−|β| ( )|α−β| = Cx0 R−|β| ( )α1 −β1 · · · ( )αn −βn
(α − β)! 2 (α − β)! 2 2
α≥β α≥β
β! R
−|β|
≤ Cx0 R−|β| Qn = Cx0 2n β! .
i=1 (1 − 12 )1+βi 2
 PDES 27

This shows that for every x0 ∈ Ω, the exist an open set Nx0 containing x0 and Mx0 > 0 and
rx0 > 0 such that
−|β|
|Dβ f (x)| ≤ Mx0 β! rx0 ∀ x ∈ Nx0 .
Let K ⊂ Ω be any compact set. Then K can be cover by finitely many open sets for which the
above estimate holds with constants Mxi > 0 and rxi > 0 . Take
M := max{Mxi }, r := min{rxi } .
Then, M > 0 and r > 0, and we get
|Dβ f (x)| ≤ M β! r−|β| , ∀ x ∈ K.
Conversely, let f ∈ C ∞ (Ω) and satisfies the given inequality. We want to show that f is real
analytic. Let x0 ∈ Ω. Choose R > 0 such that the compact set K := {x : |xi − (x0 )i | ≤ R} is
contained in Ω. Hence, by the given inequality, there exist M > 0 and r > 0 such that
|Dα f (x)| ≤ M α! r−|α| , x ∈ K. (3.4)
Let R1 > 0 such that 0 < R1 < min{R, r}. We claim that for any x ∈ K1 := {y : |yi − (x0 )i | ≤
R1 },
X Dα f (x0 )
f (x) = (x − x0 )α . (3.5)
α
α!

To prove (3.5), we proceed as follows. For any t ∈ R, define a function g(t) := f (x0 + t(x − x0 )).
One can easily check that
X |α|!
g (m) (t) = Dα f (x0 + t(x − x0 ))(x − x0 )α , m = 1, 2, · · · .
α!
|α|=m

Hence by Taylor’s expansion theorem, we have

1
g (2) (0) g (m) (0)
Z
(1) 1
f (x) = g(1) = g(0) + g (0) + + ··· + (1 − s)m g (m+1) (s) ds
2! m! m! 0
X Dα f (x0 )
= f (x0 ) + (x − x0 )α + Rm ,
α!
|α|≤m

where the remainder term Rm is given by

Z 1 X |α|!
1
Rm := (1 − s)m Dα f (x0 + s(x − x0 ))(x − x0 )α ds .
m! 0 α!
|α|=m+1

In view of (3.4), we see that

Z 1
1 X R1
|α| 1 X R1
|α|
|Rm | ≤ (1 − s)m |α|! M ds = |α|! M
m! 0 r (m + 1)! r
|α|=m+1 |α|=m+1
X R1
|α|
≤M .
r
|α|=m+1

|α|
Since Rr1 < 1, the series α Rr1
P
is convergent. Thus, we see that Rm → 0 as m → ∞ and
hence (3.5) holds true. In other words, f is real analytic at x0 . Since x0 is arbitrary, f is real
analytic in Ω. This completes the proof. 
28 A. K. MAJEE

3.1. Cauchy-Kovalevski theorem for ODE.

Theorem 3.5. Let f be a real-valued function analytic at the origin. Then the ODE
u0 (t) = f (u(t)), u(0) = 0
has a solution analytic in an open interval containing the origin.
Proof. WLOG, we may assume that u is C ∞ -solution of the above ODE. Note that
u(0) = 0, u0 (0) = f (0), u00 (0) = f 0 (u(t))u0 (t) = f 0 (0)f (0)
t=0
h i
u(3) (0) = f 00 (u(t))[f (u(t))]2 + [f 0 (u(t))]2 f (u(t)) = f 00 (0)[f (0)]2 + [f 0 (0)]2 f (0)
t=0
We see that u(3) (0) is a polynomial in 3-variables f (0), f 0 (0) and f 00 (0) with coefficients in Z+ .
Proceeding similarly, we see that
u(m) (0) = Pm f (0), f 0 (0), · · · , f (m−1) (0)

where Pm is a polynomial in m-variable with coefficients in Z+ .

Suppose there exist real valued functions g and V real analytic at the origin such that
V 0 (t) = g(V (t)), V (0) = 0; |f (m) (0)| ≤ g (m) (0) ∀ m .
V (m) (0) m
Since V is real analytic at origin, the series ∞
P
m=0 m! t has a positive radius of convergence,
say R > 0. Observe that
Pm f (0), f 0 (0), · · · , f (m−1) (0) Pm |f (0)|, |f 0 (0)|, · · · , |f (m−1) (0)|



≤
m! m!
Pm g(0), g 0 (0), · · · , g (m−1) (0)


V (m) (0)
≤ = .
m! m!


P∞ Pm f (0),f 0 (0),··· ,f (m−1) (0) m
Hence the series m=1 m! t converges absolutely in |t| < R. Define the
function u(t) as
∞
Pm f (0), f 0 (0), · · · , f (m−1) (0) m


X
u(t) = t , |t| < R .
m!
m=1
Then u is real analytic at origin and u(0) = 0. Next we show that u0 (t) = f (u(t)). Since u and
f are real analytic at origin, both u0 (t) and f (u(t)) is real analytic in an open set containing the
origin. Therefore, it is enough to show that
(u0 )(m) (0) = (f ◦ u)(m) (0), ∀ m.
This is true by the construction of u. To complete the proof, we need to show existence of g
and V .
Existence of g and V : Since f is real analytic at 0, there exists R1 > 0 such that the series
P∞ f (m) (0) m
m=0 m! R1 is convergent. Thus, there exists C > 0 such that
f (m) (0) m
R1 ≤ C =⇒ |f (m) (0)| ≤ C m! R1−m , ∀m .
m!
Define the function g as
∞
X
g(t) := CR1−m tm , |t| < R1 .
m=0
Then g is real analytic in |t| < R1 and
∞ 
X t m CR1
g(t) = C = , g (m) (0) = C m! R1−m ≥ |f (m) (0)| .
R1 R1 − t
m=0
 PDES 29

Next we want to find a real analytic function V satisfying the ODE

CR1
V 0 (t) = g(V (t)) = , V (0) = 0.
R1 − V (t)
Solving this ODE, we have
1
q
R1 V (t) − V 2 (t) = CR1 t =⇒ V (t) = R1 ± R12 − 2C R1 t .
2
Choose r
 2C  R1
V (t) := R1 1 − 1 − t , |t| < .
R1 C
Then V is real analytic in RC1 and satisfies the required property. This completes the proof. 
3.2. Majorization. One of the main ingredient in the proof of C-K theorem is to use the
method of majorization, which consists of comparing analytic functions with other functions
which have larger Taylor coefficients, but can be given explicitly.
Definition 3.2. Let Ω ⊂ Rn be open, 0 ∈ Ω and f, F : Ω → Rm be given functions. We say
that f is majorized by F , denoted by f << F , if
|Dα fk (0)| ≤ Dα Fk (0), ∀ α, k = 1, 2, · · · , m.
Example 3.2. Let f : Ω → R be a real analytic at 0 ∈ Ω ⊂ Rn . Then f << F , where F is
given by
CR
F (x) = , for some R > 0 and C > 0.
R − (x1 + x + 2 + · · · + xn )
P α f (0) |α|
Since f is real analytic at 0, there exists R > 0 such that the series α D α! R converges.
Hence, there exists C > 0 such that
Dα f (0) |α|
R ≤ C =⇒ |Dα f (0)| ≤ C α! R−|α| .
α!
Observe that
CR 1
F (x) = =C x x
1 − R + R + · · · + xRn


R − (x1 + x + 2 + · · · + xn ) 1 2

X |α|! X C α! R−|α|
=C xα R−|α| = xα .
α
α! α
α!
Thus, Dα F (0) = C α! R−|α| ≥ |Dα f (0)| for all multi-indices α.
30 A. K. MAJEE

4. Laplace Equation
We will study the most important PDE called Laplace equation resp. Poisson equation
∆u = 0, resp. − ∆u = f.
Let us recall some essential definitions.
Definition 4.1. Let Ω ⊂ Rn be a bounded open set. We say that Ω has a C k -boundary if for
each x0 ∈ ∂Ω, there exist r > 0 and a C k -function h : Rn−1 → R such that

Ω ∩ B(x0 , r) = x ∈ B(x0 , r) : xn > h(x1 , x2 , . . . , xn−1 ) .
Let f (x1 , x2 , . . . , xn ) := xn − h(x1 , x2
, . . . , xn−1 ). We define the unit outward normal at x ∈ ∂Ω,
denoted by ν(x) = ν1 (x), . . . , νn (x) as
∇f (x)
ν(x) = − .
|∇f (x)|
x−x0
Let Ω = B(x0 , R). Then for any x ∈ ∂Ω, the outward unit normal is ν(x) = R .
Let us recall integration by parts formula and Green’s theorem.
Theorem 4.1 (Integration by parts formula). Let Ω be an open bounded domain in Rn with
C 1 -boundary and u, v ∈ C 1 (Ω) ∩ C(Ω̄). Then
Z Z Z
uxi v dx = − uvxi dx + uvνi dS , (4.1)
Ω Ω ∂Ω


where ν(x) = ν1 (x), . . . , νn (x) is the outward unit normal at x ∈ ∂Ω and dS is the surface
measure on ∂Ω.
Taking v = 1 in (4.1), we get
Z Z
uxi v dx = uνi dS, u ∈ C 1 (Ω) ∩ C(Ω̄) .
Ω ∂Ω
For example, one has
xi − (x0 )i
Z Z
uxi dx = u dS .
B(x0 ,R) ∂B(x0 ,R) R
Remark 4.1. The above result is not true in general for unbounded domain. For example, take
Ω = {(x, y) ∈ R2 : y > 0}. Take u = v = 1. If the integration by parts formula holds, then one
has Z
νi dS = 0.
∂Ω
But in this case ∂Ω is x-axis.
Let Ω be a bounded open set in Rn with C 1 -boundary and u ∈ C 1 (Ω̄). Then the normal
derivative of u on ∂Ω is defined as
∂u
= ν · ∇u.
∂ν
As a consequence of integration by parts formula, the following green’s formulas hold.
Theorem 4.2 (Green’s formulas). Let Ω be a bounded open set in Rn with C 1 -boundary and
u, v ∈ C 2 (Ω̄). Then
Z Z
∂u
i) ∆u dx = dS ,
∂Ω ∂ν
ZΩ Z Z
∂v
ii) ∇u · ∇v dx = − u∆v dx + u dS ,
Ω Ω ∂Ω ∂ν
 PDES 31
Z Z
∂v ∂u

iii) (u∆v − v∆u) dx = u −v dS .
Ω ∂Ω ∂ν ∂ν
∂u
Remark 4.2. Let u ∈ C 1 (Ω̄) ∩ C 2 (Ω) and ∆u = 0 in Ω, and u = 0 or ∂ν on ∂Ω. Then from
the above theorem, we have
Z
|∇u|2 dx = 0.
Ω

Thus, u is constant in Ω̄. In the case of u = 0 on ∂Ω, u = 0 on Ω̄.

Let αn denotes the Lebesgue measure of B(0, 1) and wn−1 denotes the surface measure of
∂B(0, 1) := Sn−1 . We would like to get a relation between αn and wn−1 . To do so, we proceed
as follows. Note that for any integrable function f , we have
Z Z ∞Z 
f dx = f dS dr.
Rn 0 ∂B(0,r)

Thus, we have
Z Z ∞Z  Z 1Z 
αn = 1B(0,1) (x) dx = 1B(0,1) (x)dS(x) dr = dS dr .
Rn 0 ∂B(0,r) 0 ∂B(0,r)

By using the properties of surface measure, we note that

Z Z
n−1
dS = r dS = rn−1 wn−1 .
∂B(0,r) ∂B(0,1)

Thus, we have
Z 1
1
αn = rn−1 wn−1 dr = wn−1 =⇒ nαn = wn−1 . (4.2)
0 n
Once we calculate wn−1 , we can calculate αn by using the relation (4.2). To calculate we will
use the gamma function
Z ∞
Γ(z) = e−t tz−1 dt
0
and its properties. Observe that
Z Z ∞Z  Z ∞
−|x|2 −|x|2 2
e dx = e dS dr = e−r rn−1 wn−1 dr
Rn 0 ∂B(0,r) 0
Z ∞
wn−1 n wn−1 n
= e t 2 −1 dt =
−t
Γ( ) . (4.3)
2 0 2 2
On the other hand, we have
Z Z n Z n
−|x|2 −x2i 2
Y
e dx = e dx1 dx2 . . . dxn = e−y dy
Rn Rn i=1 R
 Z ∞
2
n 1
n n
= 2 e−y dy = Γ( ) = π 2 . (4.4)
0 2
Combining (4.3) and (4.4), we get
n
π2
wn−1 = 2 n , n ≥ 1. (4.5)
Γ( 2 )
32 A. K. MAJEE

4.1. Harmonic function and its properties.

Definition 4.2 (Harmonic function). Let Ω ⊂ Rn be an open set in Rn . We say that u ∈ C 2 (Ω)
is called a harmonic function if ∆u(x) = 0 for all x ∈ Ω.
Example 4.1. P i) u(x) = c for all x.
ii) u(x) = ni=1 ai xi , the first order polynomial.
iii) u(x, y) = x2 − y 2 is a harmonic function on R2 .
Theorem 4.3 (Mean-value theorem ). Let u ∈ C 2 (Ω) be harmonic. Then for all x ∈ Ω and
r > 0 such that B(x, r) ⊂ Ω, there holds
Z Z
1 1
u(x) = u(y) dσ(y) = u(y) dy ,
σ(∂B(x, r)) ∂B(x,r) |B(x, r)| B(x,r)
where σ is the surface measure.
Proof. Fix x ∈ Ω and define r > 0 such that B(x, r) ⊂ Ω. Define
Z
1
φ(r) := u(y) dσ(y).
σ(∂B(x, r)) ∂B(x,r)
Then, one has
Z
1
φ(r) = u(x + y) dσ(y)
σ(∂B(0, r)) ∂B(0,r)
Z
1
= u(x + rz) dσ(z)
σ(∂B(0, r)) ∂B(0,1)
Z
1
= u(x + rz) dσ(z) (as σ(∂B(0, r)) = rn−1 σ(∂B(0, 1))) .
σ(∂B(0, 1)) ∂B(0,1)
Differentiating under the integral sign, we have
Z
0 1
φ (r) = ∇u(x + rz) · z dσ(z)
σ(∂B(0, 1)) ∂B(0,1)
y−x
Z
1
= ∇u(y) · dσ(y) .
σ(∂B(x, r)) ∂B(x,r) r
Since y−x ∂u
r is the normal to the surface ∂B(x, r) at a point y and ν · ∇u = ∂ν , by using Green’s
formula, we have
Z
0 1
φ (r) = ∆u(y) dy = 0 (as u is harmonic).
σ(∂B(x, r)) B(x,r)
This shows that φ(r) is constant and hence φ(r) = lims→0 φ(s). Now φ(s) can be re-written as
Z
1 
φ(s) = u(x + sz) − u(x) + u(x) dσ(z)
wn−1 Sn−1
Z
1 
= u(x + sz) − u(x) dσ(z) + u(x) .
wn−1 Sn−1
Note that for z ∈ Sn−1 , |(x + sz) − x| = |sz| = s. Since u is continuous, given  > 0, there exists
δ > 0 such that
|u(x + sz) − u(x)| <  for s ≤ δ.
Hence, for all s ≤ δ, Z
1 
u(x + sz) − u(x) dσ(z) < .
wn−1 Sn−1
 PDES 33

This shows that

Z
1
u(y) dσ(y) = φ(r) = lim φ(s) = u(x). (4.6)
σ(∂B(x, r)) ∂B(x,r) s→0

Next we prove the second equality. Note that for any locally integrable function f , one has
Z Z r Z 
f (y) dy = f (y) dσ(y) ds .
B(x,r) 0 ∂B(x,s)
Taking f = 1 in the above equality, one has
Z r
|B(x, r)| = σ(∂B(x, s)) ds.
0
We now have
Z Z r Z 
u(y) dy = f (y) dσ(y) ds
B(x,r) 0 ∂B(x,s)
Z r Z
 n 1 o
= σ(∂B(x, s)) f (y) dσ(y) ds
0 σ(∂B(x, s)) ∂B(x,s)
Z r
= σ(∂B(x, s))u(x) ds (by (4.6))
0
= u(x)|B(x, r)|
Z
1
=⇒ u(x) = u(y) dy .
|B(x, r)| B(x,r)
This completes the proof. 
Theorem 4.4. If u ∈ C 2 (Ω) satisfies the mean value theorem i.e.,
Z
1
u(x) = u(y) dσ(y)
σ(∂B(x, r)) ∂B(x,r)
for each ball B(x, r) ⊂ Ω, then u is harmonic in Ω.
Proof. For fixed x ∈ Ω, let r > 0 such that B(x, r) ⊂ Ω. Define
Z
1
φ(r) := u(y) dσ(y).
σ(∂B(x, r)) ∂B(x,r)
Then we have seen that Z
0 1
φ (r) = ∆u(y) dy.
σ(∂B(x, r)) B(x,r)
From the given condition, φ(r) = u(x) for all r such that B(x, r) ⊂ Ω. Hence φ0 (r) = 0. Thus ,
we have
Z Z
1
∆u(y) dy = 0 =⇒ ∆u(y) dy = 0 .
B(x,r) |B(x, r)| B(x,r)
Observe that
Z
1
∆u(y) dy − ∆u(x)
|B(x, r)| B(x,r)
Z
1
≤ |∆u(y) − ∆u(x)| dy → 0 as r → 0 (by continuity of ∆u)
|B(x, r)| B(x,r)
Z
1
=⇒ ∆u(x) = lim ∆u(y) dy = 0
r→0 |B(x, r)| B(x,r)

=⇒ ∆u(x) = 0 ∀ x ∈ Ω =⇒ u is harmonic in Ω.
34 A. K. MAJEE


Corollary 4.5. If u ∈ C 2 (Ω) satisfies ∆u ≥ 0 in Ω, then
Z
1
u(x) ≤ u(y) dσ(y) ∀ B(x, r) ⊂ Ω .
σ(∂B(x, r)) ∂B(x,r)
Moreover, one has Z
1
u(x) ≤ u(y) dy ∀ B(x, r) ⊂ Ω .
|B(x, r)| B(x,r)
1
R
Proof. Let φ(r) := σ(∂B(x,r)) ∂B(x,r) u(y) dσ(y). Then we have proved that
Z
0 1
φ (r) = ∆u(y) dy.
σ(∂B(x, r)) B(x,r)
Since ∆u ≥ 0 in Ω, this shows that φ0 (r) ≥ 0 and hence
Z
1
u(x) = lim φ(s) ≤ φ(r) = u(y) dσ(y).
s→0 σ(∂B(x, r)) ∂B(x,r)
To prove the second inequality, we observe that, since u(x) ≤ ψ(s) for s > 0,
Z Z r Z
n 1 o
u(y) dy = σ(∂B(x, s)) f (y) dσ(y) ds
B(x,r) 0 σ(∂B(x, s)) ∂B(x,s)
Z r Z r
= sn−1 wn−1 ψ(s) ds ≥ sn−1 wn−1 u(x) ds = u(x)rn αn = u(x)|B(x, r)|
0 0
Z
1
=⇒ u(x) ≤ u(y) dy ∀ B(x, r) ⊂ Ω .
|B(x, r)| B(x,r)

Theorem 4.6. Let u ∈ C(U ), and satisfies the mean-value property (MVT) i.e.,
Z
1
u(x) = u(y) dσ(y)
σ(∂B(x, r)) ∂B(x,r)
for all B(x, r) ⊂ U . Then u ∈ C ∞ (U ).
1 x
Proof. Let φ = n φ(  ) be the standard mollifier. Denote the set
Ω := {x ∈ Ω : dist(x, ∂Ω) > }.
For any x ∈ Ω , we have
Z Z
x−y
Z
1 
u ∗ φ (x) = u(y)φ (x − y) dy = n u(y)φ( ) dσ(y) dr .
B(x,)  0 ∂B(x,r) 
Since φ is radial function, it depends only on |x| not on x. Thus, we have
1  r 
Z Z 
u ∗ φ (x) = n φ( ) u(y), dσ(y) dr
 0  ∂B(x,r)
Z 
1 r
= n φ( )u(x)σ(∂B(x, r)) dr (by MVT)
 0 
Z 
= u(x) φ (r)σ(∂B(x, r)) dr .
0
Observe that
Z Z Z 
1= φ (y) dy = ψ (z) dσ(z) dr
B(0,) 0 ∂B(0,r)
 PDES 35
Z  Z 
= φ (r)σ(∂B(0, r)) dr = φ (r)σ(∂B(x, r)) dr .
0 0

Thus, we get
u ∗ φ = u in Ω .
Since u ∗ φ ∈ C ∞ (Ω ), and  is arbitrary, we get u ∈ C ∞ (Ω). 
Remark 4.3. The above theorems imply that if u satisfies MVT, then it is harmonic and
u ∈ C ∞ . If u is harmonic, then it satisfies MVT and hence u ∈ C ∞ .
More generally, the following theorem is due to Weyl.
Theorem 4.7 (Weyl). Let u : Ω → R be measurable and locally integrable in Ω. If u satisfies
∆u = 0 in D0 (Ω), in the sense of distribution, then u is harmonic and u ∈ C ∞ (Ω).
We have the following estimate on the derivatives of a harmonic function.
Theorem 4.8 (Estimates on derivatives). Let u be a harmonic function in Ω and α is a multi-
index with |α| = k. Then
Ck
|Dα u(x0 )| ≤ n+k kukL1 (B(x0 ,r))
r
for each ball B(x0 , r) ⊂ Ω, where
1 (2n+1 nk)k
C0 = , Ck = , k = 1, 2, . . . .
αn αn
Proof. We prove this result by induction. Let |α| = 0. Then by MVT, we have
Z
1 1
u(x0 ) = u(y) dy ≤ kukL1 (B(x0 ,r)) .
|B(x0 , r)| B(x0 ,r) αn r n
Thus, the assertion holds for k = 0. To prove the assertion for k = 1, we proceed as follows.
Since u is harmonic, uxi is also harmonic and hence satisfies the MVT. Thus, by using green’s
theorem, we get
Z Z
1 1
uxi (x0 ) = ux (y) dy = u(y)νi (d) dσ(y)
|B(x0 , 2r )| B(x0 , r ) i |B(x0 , 2r )| ∂B(x0 , r )
2 2
Z
1 1 r
=⇒ |uxi (x0 )| ≤ r |u(y)| dσ(y) ≤ r sup |u(y)|σ(∂B(x0 , ))
|B(x0 , 2 )| ∂B(x0 , r ) |B(x0 , 2 )| y∈∂B(x0 , r ) 2
2 2

wn−1 ( 2r )n−1 2wn−1

= sup |u(y)| = sup |u(y)| .
αn ( 2r )n y∈∂B(x0 , r ) αn r y∈∂B(x0 , r2 )
2

Fix y0 ∈ ∂B(x0 , 2r ). Then B(y0 , 2r ) ⊂ B(x0 , r) ⊂ Ω. We get

Z Z
1 1 1 1
|u(y0 )| ≤ |u(y)| dy ≤ |u(y)| dy
αn ( 2r )n B(y0 , r ) αn ( 2r )n B(x0 ,r)
2
Z
1 1
=⇒ sup |u(y)| ≤ |u(y)| dy .
y∈∂B(x0 , r ) αn ( 2r )n B(x0 ,r)
2

Thus, we have, using the fact that nαn = wn−1

2wn−1 1 1 C1
|uxi (x0 )| ≤ kukL1 (B(x0 ,r)) = n+1 kukL1 (B(x0 ,r)) .
αn r αn ( 2r )n r
36 A. K. MAJEE

Thus the assertion holds for k = 1. Now assume k ≥ 2 and the assertion holds for all |α| ≤ k − 1.
Let |α| = k. Then dα u = (Dβ u)xi for some i ∈ {1, . . . , n} and |β| = k−1. Since Dα u is harmonic,
we have
Z Z
1 1
Dα u(x0 ) = (D β
u) xi dy = (Dβ u)νi dσ(y)
|B(x0 , kr )| B(x0 , r ) |B(x0 , kr )| ∂B(x0 , r )
k k

αnk β
=⇒ |D u(x0 )| ≤ kD ukL∞ (B(x0 , kr )) .
r
If y ∈ B(x0 , kr ), then B(y, k−1
k r) ⊂ B(x0 , r). Thus, by induction hypothesis, we have
Z Z
β Ck−1 Ck−1
|D u(y)| ≤ k−1 |u(z)| dz ≤ k−1 |u(z)| dz
( k r)n−k+1 B(y, k−1
k
r) ( k r)n−k+1 B(x0 ,r)
Thus, combining last two inequality, we get the assertion for |α| = k. this completes the
proof. 
Theorem 4.9 (Liouville’s theorem). Any bounded harmonic function on Rn is a constant.
Proof. Let u be a harmonic on Rn . Then for any x ∈ Rn , and r > 0,
Z
C1 C1
|uxi (x)| ≤ n+1 |u(y)| dy ≤ sup |u| n+1 rn |B(0, 1)| → 0 as r → ∞ .
r B(x,r) r
Therefore uxi (x) = 0. Since x is arbitrary, we see that u is constant. 
Theorem 4.10. Let u be a harmonic function in Ω. Then u is real analytic.
Proof. Let K ⊂ Ω be compact. We want to show that there exist M > 0, r > 0 such that
|Dα u(x)| ≤ M α!r−|α| ∀ x ∈ K, ∀ α. (4.7)
Choose r > 0 such that 0 < r < dist(K, ∂Ω). Then for all x ∈ K, B(x, r) ⊂ Ω. Hence by the
derivative estimate of harmonic function,
Z
α Ck
|D u(x)| ≤ n+k |u(y)| dy ∀ x ∈ K, ∀ α.
r B(x,r)

Let K̃ := {x ∈ Ω : dist(x, K) ≤ r}. Then for all x ∈ K, one has B(x, r) ⊂ K̃ and K̃ is compact.
Thus, we have
2n+1 n k
Z
α Ck
|D u(x)| ≤ n+k |u(y)| dy = M k k
r K̃ r
R
|u(y)| dy kk
where M := K̃
αn rn . Since ek > k! , we have

2n+1 n k −|α|
|Dα u(x)| ≤ M k!ek ≤ M |α|!r1 .
r
Hence u is real analytic in Ω. 

We now deduce some important properties of subharmonic function.

Theorem 4.11 (Weak maximum principle). Let Ω be a bounded domain in Rn and u ∈ C 2 (Ω)∩
C(Ω̄) satisfies ∆u ≥ 0 in Ω. Then
max u = max u.
Ω̄ ∂Ω
 PDES 37

Proof. Case 1: ∆u > 0 in Ω. If x0 is a point of interior maximum of u, then by second deriv-

2
ative test, ∂∂xu2 (x0 ) ≤ 0–this then shows that ∆u(x0 ) ≤ 0—-a contradiction.
i
Case 2: ∆u ≥ 0 in Ω. For ε > 0, consider the function
vε (x) := u(x) + ε|x|2 , x ∈ Ω.
Then vε ∈ C 2 (Ω) ∩ C(Ω̄) and ∆vε (x) = ∆u + 2nε > 0 in Ω. Hence by case 1, vε attains its
maximum only on the boundary ∂Ω i.e.,
max v = max v .
Ω̄ ∂Ω

Observe that
max u +  min |x|2 ≤ max u(x) + |x|2 = max v (x) = max v (x) ≤ max u(x) +  max |x|2 .


Ω̄ Ω̄ Ω̄ Ω̄ ∂Ω ∂Ω ∂Ω

Taking  → 0, we arrive at the conclusion. 

Corollary 4.12. If u ∈ C 2 (Ω) ∩ C(Ω̄) satisfies −∆u ≥ 0 in Ω and u ≥ 0 on ∂Ω, then u ≥ 0 in
Ω.
Proof. Let v = −u. Then ∆v ≥ 0 in Ω and v ≤ 0 in ∂Ω. Hence by weak maximum principle,
we have
max v = max v ≤ 0 =⇒ v ≤ 0 in Ω =⇒ u ≥ 0 in Ω .
Ω̄ ∂Ω


Remark 4.4. In case of harmonic function i.e., ∆u = 0, Theorem 4.11 holds for −u as well.
Therefore, using the fact that min u(x) = − max(−u(x)), we obtain
min u = min u .
Ω̄ ∂Ω

Since |a| = max{a, −a}, we have, for harmonic function u with u ∈ C(Ω̄)
max |u| = max |u| .
Ω̄ ∂Ω

Example 4.2. Let u(x, y) = x2 − y 2 . Then u ∈ C 2 (U ) ∩ C(Ū ) where U = {(x, y) : x2 + y 2 < 1}.
Moreover, u is harmonic, and max u = 1 which attained at boundary points (1, 0), (−1, 0).
Ū
Furthermore, min u = −1 which attains again at the boundary points (0, 1) and (0, −1).
Ū

Theorem 4.13 (Uniqueness). Let g ∈ C(∂U ), f ∈ C(U ) and k ≥ 0. Then there exists at most
one solution u ∈ C 2 (U ) ∩ C(Ū ) of the BVP
∆u − ku = f in U ; u = g on ∂U .

Proof. Suppose that u, v ∈ C 2 (U ) ∩ C(Ū ) solves the BVP. Define w = u − v. Then w ∈

C 2 (U ) ∩ C(Ū ) and solves the homogeneous Dirichlet problem
∆w − kw = 0 in U ; w = 0 on ∂U .
Multiplying the above equation by w, and use integration by parts formula, we have
Z Z
2
− |∇w| − k |w|2 = 0.
Ω Ω

Since k ≥ 0, we have w is constant. Since w = 0 on ∂U , and w ∈ C(Ū ), we have w = 0 i.e.,

u = v. This finishes the proof. 
38 A. K. MAJEE

Remark 4.5. For unbounded domains, the above theorem may not be true. For example,
consider the following problem on the upper half plane

∆u(x, y) = 0 for x ∈ R, 0 < y < ∞; u(x, 0) = 0 ∀ x ∈ R .

This problem has at least two solutions, namely u1 (x, y) = xy and u2 (x, y) = 0.

Example 4.3. Let u ∈ C 2 (Ω) ∩ C(Ω̄) be a solution of the PDE

−∆u = f in Ω, u = g on ∂Ω

where f ∈ C(Ω̄) and g ∈ C(∂Ω). Then the following inequality holds:

 
sup |u(x)| ≤ C sup |g(y)| + sup |f (x)| .
x∈Ω y∈∂Ω x∈Ω

Solution: Define
|x|2
v(x) := u(x) + sup |f (y)|, x ∈ Ω.
2n y∈Ω

Then v ∈ C 2 (Ω) ∩ C(Ω̄) and ∆v(x) = −f (x) + supy∈Ω |f (y)| ≥ 0. Hence by weak maximum
principle, maxΩ̄ v = max∂Ω v. Hence, one has, for x ∈ Ω

|x|2 1  2
u(x) + sup |f (y)| = v(x) ≤ max v(x) = max v(y) ≤ sup |u(x)| + sup |x| sup |f (x)| .
2n y∈Ω Ω̄ ∂Ω ∂Ω 2n x∈Ω̄ x∈Ω

Since u = g on ∂Ω, we get

n o
u(x) ≤ C(Ω, n) sup |g(y)| + sup |f (x)| .
y∈∂Ω x∈Ω

Theorem 4.14 (Strong maximum principle). Let Ω be a connected domain and let u ∈ C 2 (Ω)
satisfies ∆u ≥ 0 in Ω. Then either u is constant or u(ξ) < supx∈Ω u(x) for all ξ ∈ Ω.

Proof. Let A := supx∈Ω̄ u(x) < ∞. Then by continuity of u, the set M := {x ∈ Ω : u(x) = A}
is relatively closed set in Ω. We now show that M is open in Ω. Suppose there exists x0 ∈ Ω
such that u(x0 ) = A i.e., M is non-empty. Then for 0 < r < dist(x0 , ∂Ω), by Corollary 4.5
Z
1
A = u(x0 ) ≤ u(y) dy
|B(x0 , r)| B(x0 ,r)
Z Z


=⇒ 0 ≤ u(y) dy − |B(x0 , r)|A = u(y) − A dy.
B(x0 ,r) B(x0 ,r)

But u(y) − A ≤ 0 and u is continuous. Hence there exists r > 0 such that u(y) = A for all
y ∈ B(x0 , r). In other words, M is open in Ω. Since Ω is connect, we have either M is empty or
M = Ω i.e., either u is constant or u(ξ) < supx∈Ω u(x) for all ξ ∈ Ω. 

Remark 4.6. The second part of Theorem 4.11 asserts in particular that if U is connected and
u ∈ C 2 (U ) ∩ C(Ū ) satisfies

−∆u = 0 in U ; u = g on ∂U ,

where g ≥ 0, then u > 0 in U provided g(y) > 0 for some y ∈ ∂U .

 PDES 39

4.2. Fundamental solution of Laplace equation. Let us find a solution u of the Laplace
∂r
equation in Rn of the form u(x) = v(r), r = |x|. Note that ∂xi
= xri . Thus,

xi x2 1 x2
n−1 0
uxi = v 0 (r) , uxi xi = v 00 (r) 2i + v 0 (r) − 3i =⇒ ∆u = v 00 (r) + v (r) .
r r r r r
Hence
n−1 0
0 a
∆u = 0 ⇔ v 00 (r) + v (r) =⇒ rn−1 v 0 = 0 =⇒ v 0 = for some constant a .
r rn−1
Consequently, if r > 0, we have
(
b log(r) + c, (n = 2)
v(r) = b
rn−2
+ c, (n ≥ 3) ,
where b and c are constants. These considerations motivate the following definition:
Definition 4.3. The function
(
1
− 2π log(|x|) (n = 2)
Φ(x) := 1 1 (4.8)
n(n−2)αn |x|n−2 (n ≥ 3)

defined for x ∈ Rn , x 6= 0, is called the fundamental solution of Laplace equation.

Remark 4.7. The fundamental solution of Laplace equation Φ satisfies the following properties:
i) Φ, ∇Φ are locally integrable. One has
(
1 x
− 2π |x|2
, n=2 1 x
∇Φ(x) = 1 x =⇒ ∇Φ(x) = − for any n ≥ 2 .
− nαn |x|n , n ≥ 3 wn−1 |x|n
∂2Φ
ii) is Not locally integrable. Moreover, x 7→ Φ(x) is harmonic for x 6= 0, and hence
∂xi ∂xj
x 7→ Φ(x − y) is harmonic as a function of x, x 6= y.
Let us recall the definition of convolution of two function. We define
Z
(f ∗ g)(x) := f (y)g(x − y) dy
Rn
whenever the integral on the right hand side exists. It is easy to see that
f ∗ g = g ∗ f.
If f ∈ L1 (Rn ) and g ∈ L∞ (Rn ), then f ∗ g ∈ L∞ (Rn ). Indeed,
Z Z
|(f ∗ g)(x)| ≤ |f (y)||g(x − y)| dy ≤ kgkL∞ (Rn ) |f (y)| dy = kgkL∞ (Rn ) kf kL1 (Rn ) < +∞ .
Rn Rn

If f ∈ L1 (Rn ) and g ∈ C 1 (Rn )

with |g 0 |
≤ M for some M > 0, then f ∗ g ∈ C 1 (Rn ) and
(f ∗ g)0 = f ∗ g 0 . In general in Rn , one has
Dα (f ∗ g) = f ∗ Dα g.
Theorem 4.15. Let f ∈ Cc2 (Rn ). Then u = Φ ∗ f ∈ C 2 (Rn ) and solves −∆u = f in Rn .
Proof. Since f ∈ Cc2 (Rn ), u = φ ∗ f is well-defined. Now
u(x + hei ) − u(x) f (x + hei − y) − f (x − y)
Z
= Φ(y) dy .
h Rn h
40 A. K. MAJEE

∂u ∂u ∂f
Hence, by Lebesgue dominated convergence theorem, ∂xi exists and ∂xi = ∂xi ∗ Φ. In the
∂2u ∂2u ∂2f ∂2f
same way, ∂xi ∂xj exists and ∂xi ∂xj = Φ ∗ ∂xi ∂xj . Since ∂xi ∂xj (x) is continuous and Φ is locally
2u
integrable, one can easily check that ∂x∂i ∂x j
is continuous. Hence u ∈ C 2 (Rn ). Now
Z Z
∆u(x) = Φ(y)∆x f (x − y) dy + Φ(y)∆x f (x − y) dy := I1ε + I2ε .
B(0,ε) Rn \B(0,ε)
Observe that
(
Cε2 | log(ε)|, n = 2
Z
|I1ε | ≤C |Φ(y)| dy ≤
B(0,ε) Cε2 , n ≥ 3 .
Thus, I1ε → 0 as ε → 0. By using integration by parts formula, we see that
Z Z
∂f
I2ε = − DΦ(y) · Dy f (x − y) dy + Φ(y) (x − y) dσ(y)
Rn \B(0,ε) ∂B(0,ε) ∂ν
ε ε
= I2,1 + I2,2 ,
where ν is the unit normal along ∂B(0, ε). Notice that
Z (
ε Cε| log(ε)|, n = 2
|I2,2 |≤C |Φ(y)| dσ(y) ≤
∂B(0,ε) Cε, n ≥ 3
ε → 0 as ε → 0. Again, by using integration by parts formula, we have
and hence I2,2
Z Z
ε ∂Φ
I2,1 = ∆Φ(y)f (x − y) dy − (y)f (x − y) dσ(y)
Rn \B(0,ε) ∂B(0,ε) ∂ν
Z
∂Φ
=− (y)f (x − y) dσ(y) .
∂B(0,ε) ∂ν

Note that ∂Φ
∂ν = ν · DΦ(y), and ν = − yε on ∂B(0, ε), and DΦ(y) = − nαny|y|n . Consequently,
Z Z
ε 1 1
I2,1 =− n−1
f (x − y) dσ(y) = − f (y)dσ(y)
∂B(0,ε) nαn ε σ(∂B(x, ε)) ∂B(x,ε)
→ −f (x) as ε → 0.
Thus, −∆u = f in Rn . 
Remark 4.8. Theorem 4.15 holds true for f ∈ Cc1 (Rn ).
4.3. Green’s function: We are interested in finding a general representation formula for the
solution of Poisson’s equation
−∆u = f in U, u = g on ∂U , (4.9)
where U is a bounded domain with C 1 boundary. Let Φ be a fundamental solution. Fix x ∈ U .
Let u ∈ C 2 (U ) ∩ C 1 (Ū ). Consider the function y 7→ Φ(y − x). Then by integration by parts
formula
Z Z
∂u ∂Φ
∆u(y)Φ(y − x) dy = [ (y)Φ(y − x) − u(y) (y − x)] dσ(y) .
U \B(x,ε) ∂U ∪∂B(x,ε) ∂ν ∂ν
R
Taking ε → 0, the l.h.s becomes U ∆u(y)Φ(y − x) dy. Now, r.h.s can be written as
Z
∂u ∂Φ
r.h.s. = [ (y)Φ(y − x) − u(y) (y − x)] dσ(y)
∂U ∂ν ∂ν
Z
∂u ∂Φ
+ [ (y)Φ(y − x) − u(y) (y − x)] dσ(y)
∂B(x,ε) ∂ν ∂ν
 PDES 41
Z
∂u ∂Φ
→ [ (y)Φ(y − x) − u(y) (y − x)] dσ(y) − u(x) .
∂U ∂ν ∂ν
Thus, we have
Z Z
∂u ∂Φ
u(x) = − ∆u(y)Φ(y − x) dy + [ (y)Φ(y − x) − u(y) (y − x)] dσ(y) . (4.10)
U ∂U ∂ν ∂ν
Suppose that u solves the Poisson’s equation, then
Z Z
∂u ∂Φ
u(x) = f (y)Φ(y − x) dy + [ (y)Φ(y − x) − g(y) (y − x)] dσ(y) .
U ∂U ∂ν ∂ν
∂u
Therefore, the normal derivative along ∂U is unknown to us. This motivates to find a
∂ν
corrector function Φx (y) solving the boundary value problem
∆Φx (y) = 0 in U , Φx (y) = Φ(y − x) on ∂U .
Then by integration by parts formula, we have
 ∂Φx (y)
Z Z
x ∂u 
− Φ (y)∆u(y) dy = u − Φx (y) (y) dσ(y)
U ∂ν ∂ν
Z∂U x
 ∂Φ (y) ∂u 
= u − Φ(y − x) (y) dσ(y) . (4.11)
∂U ∂ν ∂ν
Combining (4.10) and (4.11), we get, if u solves the Poisson’s equation,
Z Z
x ∂ 
Φ(y − x) − Φx (y) dσ(y) .
  
u(x) = f (y) Φ(y − x) − Φ (y) dy − g(y) (4.12)
U ∂U ∂ν
Definition 4.4 (Green’s function). G(x, y) = Φ(y − x) − Φx (y), x, y ∈ U, x 6= y is called the
Green’s function of the domain U .
Remark 4.9. Φx (y) depends upon the domain U .
Properties of Green’s function: Following properties of the Green’s function G hold:
a) ∆y G(x, y) = 0 = ∆x G(x, y) for all x, y ∈ U with y 6= x.
b) G(x, y) = G(y, x) for all x, y ∈ U with y 6= x.
c) G(x, y) > 0 for all x, y ∈ U with y 6= x.
Theorem 4.16. If u ∈ C 2 (Ū ) solves the Poisson’s equation (4.9), then
Z Z
∂G
u(x) = f (y)G(x, y) dy − g(y) (x, y) dσ(y) , (4.13)
U ∂U ∂ν
where G(x, y) is a Green’s function.
4.3.1. Green’s function for R+ x
n = {(x1 , x2 , . . . , xn ) : xn > 0}: We want to find Φ (y) such
that ∆Φx (y) = 0, y ∈ R+ x +
n and Φ (y) = Φ(y − x) for y ∈ ∂Rn . For any x ∈ Rn , define
+

x̄ = (x1 , x2 , . . . , −xn ). Define

Φx (y) = Φ(y − x̄).
It is easy to show that ∆Φx (y) = 0 for all y ∈ R+ +
n . Note that, for any y ∈ ∂Rn , |y − x| = |y − x̄|.
Since Φ is radial function, we have Φ(y − x) = Φ(y − x̄). In other words, Φx (y) = Φ(y − x) for
all y ∈ ∂R+ +
n . Thus, the Green’s function for Rn = {(x1 , x2 , . . . , xn ) : xn > 0} is given by
G(x, y) = Φ(y − x) − Φ(y − x̄), x, y ∈ R+
n , x 6= y . (4.14)
One can easily show that
∂G 1  yn − x n yn + xn 
=− n
− .
∂yn nαn |x − y| |y − x̄|n
42 A. K. MAJEE

Consequently, if y ∈ ∂R+
n,
∂G ∂G 2xn 1
(x, y) = − (x, y) = − .
∂ν ∂yn nαn |x − y|n
Theorem 4.17. Assume that g ∈ C(Rn−1 ) ∩ L∞ (Rn−1 ). Then the function defined by
Z
u(x) = K(x, y)g(y) dy,
Rn−1
where
2xn 1 + n−1
K(x, y) =
n , x ∈ Rn , y ∈ R ,
nαn x2 + Qn−1 (xi − yi )2 2
n i=1
satisfies the following:
i) u ∈ C ∞ (R+ ∞ +
n ) ∩ L (Rn ).
+
ii) ∆u = 0 in Rn .
iii) lim u(x) = g(y) for each point y ∈ ∂R+
n.
x→y,x∈R+
n

Remark 4.10. The function K(x, y) is called the Poisson kernel for R+
n and the function u is
+
called the Poisson’s formula for Rn .
Example 4.4. Let n = 2, and U be the half plane x2 > 0. Suppose that g ∈ C(R) ∩ L∞ (R).
Then the function defined by
1 ∞
Z
x2 g(y1 )
u(x1 , x2 ) = dy1 .
π −∞ x22 + (x1 − y1 )2
solves the Dirichlet problem ∆u = 0 in R+ 2 and u(x, 0) = g(x) on R. Moreover, u is bounded.
Indeed
Z
C 1 C
|u(x1 , x2 )| ≤ dy ≤ .
π R y2 π
x
4.3.2. Green’s function for ball B(0, 1): Let x ∈ B. Define x̃ = |x|2
. x̃ is called the inversion
w.t.to the unit sphere. Define
Φx (y) = Φ(|x|(y − x̃)) .
Clearly, Φx (y) is harmonic in B. Suppose y ∈ ∂B. We claim that Φx (y) = Φ(y − x). Indeed,
for any y ∈ ∂B
2y · x 1

|x|2 |y − x̃|2 = |x|2 |y|2 − 2
+ 2 = |x|2 − 2y · x + 1 = |x − y|2
|x| |x|
=⇒ Φ(|x||y − x̃|) = Φ(|y − x|) = Φ(y − x) .
Therefore, the Green’s function for B is
G(x, y) = Φ(y − x) − Φ(|x|(y − x̃)) .
2
Note that ∂G(x,y)
∂ν
1−|x|
= ∇y G(x, y) · y = − nα1 n |x−y| 2
n . Suppose that u ∈ C (B) ∩ C(B̄) solves the

Poisson’s equation. Then we have

1 − |x|2
Z
dσ(y)
u(x) = g(y) .
nαn ∂B |x − y|n
Suppose v satisfies the Poisson’s equation on B(0, R) for R > 0. Define
w(x) = V (Rx), x ∈ B .
 PDES 43

Then w satisfies the Poisson’s equation in B with boundary condition g̃(x) = g(Rx), x ∈ ∂B.
2 R
Therefore, w(x) = 1−|x|
nαn
dσ(y)
∂B g(Ry) |x−y|n . Putting Ry = x, we have R
n−1 dσ(y) = dσ(z), and

hence
1 − |x|2 R(1 − |x|2 )
Z Z
dσ(z) g(z)
V (Rx) = g(z)R n
= n
dσ(z)
nαn ∂B(0,R) |Rx − z| nαn ∂B(0,R) |Rx − z|
R2 − |X|2
Z Z
g(z)
=⇒ V (X) = n
dσ(z) = K(X, z)g(z) dσ(z)
nαn R ∂B(0,R) |X − z| ∂B(0,R)
where
R2 − |X|2 1
K(X, z) = , x ∈ B(0, R) , z ∈ ∂B(0, R) .
nαn R |X − z|n
R
K(x, z) is called Poisson’s kernel for B(0, R) and u(x) = ∂B(0,R) K(x, z)g(z) dσ(z) is called the
Poisson’s formula for the ball B(0, R).
R
Theorem 4.18. Let g ∈ C(∂B(0, R)), and define u(x) = ∂B(0,R) K(X, z)g(z) dσ(z). Then
i) u is harmonic in B(0, R).
ii) u is bounded. More precisely, |u(x)| ≤ kgkL∞ .
iii) lim u(x) = g(y) for each point y ∈ ∂B(0, R).
x→y,x∈B(0,R)

Example 4.5. Let u be positive and harmonic in B(0, R). Then show that
R − |x| R + |x|
Rn−2 n−1
u(0) ≤ u(x) ≤ Rn−2 u(0) . (4.15)
(R + |x|) (R − |x|)n−1
Solution: Suppose u solves the Poisson equation with g > 0. Then, by maximum principle u is
positive and harmonic in B(0, R). Moreover, u has the explicit form
R2 − |x|2
Z
1
u(x) = g(z) dσ(z).
nαn R ∂B(0,R) |x − z|n
Note that
R2
Z Z
1 2−n 1
u(0) = g(z) n dσ(z) = R g(z) dσ(z)
nαn R ∂B(0,R) |z| nαn R ∂B(0,R)
Z
1
=⇒ g(z) dσ(z) = Rn−2 u(0) .
nαn R ∂B(0,R)
By triangle inequality, we have R − |x| ≤ |x − z| for all z ∈ ∂B(0, R). Since g > 0, we have
(R − |x|)(R + |x|)
Z
1
u(x) = g(z) dσ(z)
nαn R ∂B(0,R) |x − z|n
(R − |x|)(R + |x|)
Z
1
≤ g(z) dσ(z)
nαn R ∂B(0,R) (R − |x|)n
R + |x| R + |x|
Z
1
≤ n−1
g(z) dσ(z) = Rn−2 u(0) .
(R − |x|) nαn R ∂B(0,R) (R − |x|)n−1
R−|x|
Similarly, using |x − z| ≤ |x| + R for all z ∈ ∂B(0, R), we have Rn−2 (R+|x|)n−1 u(0) ≤ u(x). The

inequality (4.15) is called Harnack’s inequality.

Example 4.6. Let u be a non-negative harmonic function on R2 . Then u is constant.
Solution: By Harnack’s inequality, for any R > 0, we have
R − |x| R + |x|
u(0) ≤ u(x) ≤ u(0) .
(R + |x|) (R − |x|)
44 A. K. MAJEE

Sending R → ∞, we get u(x) = u(0), i.e., u is constant.

4.4. Dirichelet’s principle: A solution of the Poisson’s problem can be characterised as the
minimizer of energy function. Define X := {u ∈ C 2 (U ) ∩ C(Ū ) : u = g on ∂U }, and J : X → R
by Z Z
1 2
J(u) = |∇u| − fu .
2 U U

Theorem 4.19. u ∈ C 2 (U ) ∩ C(Ū ) is a solution of the Poisson equation (4.9) if and only if
J(u) = min J(v).
v∈X

Proof. Suppose that J(u) = min J(v). Fix φ ∈ Cc∞ (U ). Define B : R → R by B(t) = J(u + tφ).
v∈X
d
The scalar function B(·) has a minimum at t = 0. Hence dt J(u + tφ) = 0. Now
t=0
t2
Z Z Z Z Z
1 2 2
J(u + tφ) = |∇u| + t ∇u · ∇φ + |∇φ| − fu − t fφ
2 U U 2 U
Z Z Z U U
d
=⇒ 0 = J(u + tφ) = ∇u · ∇φ − f φ = − (∆u + f )φ, ∀φ ∈ Cc∞ (U ) .
dt t=0 U U U
Thus, −∆u = f in U . Since u ∈ X, u = g on ∂U .
Conversely, suppose u solves the Poisson equation. Let v ∈ X. We want to show that
J(u) ≤ J(v). Since −∆u = f , we get
Z Z Z Z Z
− ∆u(v − u) = f (v − u) =⇒ ∇u · ∇(v − u) = fv − fu
U U U U U
Z Z Z Z Z Z Z
1 1
=⇒ |∇u|2 − fu = ∇u · ∇v − fv ≤ |∇u|2 + |∇v|2 − fv
U U U U 2 U 2 U U
Z Z Z Z
1 1
=⇒ |∇u|2 − fu ≤ |∇v|2 − f v i.e., J(u) ≤ J(v) .
2 U U 2 U U
Since v ∈ X is arbitrary, and u ∈ X, we conclude that J(u) = min J(v). 
v∈X
 PDES 45

5. Heat Equation
We will study the heat equation resp. non homogeneous heat equation

ut − ∆u = 0, resp. ut − ∆u = f

subject to appropriate initial and boundary conditions.

Example 5.1. Suppose u is smooth and solves the heat equation ut − ∆u = 0 in Rn × (0, ∞).
Then show that v(x, t) = x · Du + 2tut solves the heat equation.
Solution: Note that , since ut = ∆u, we have

vt = x · Dut + 2ut + 2tutt = x · D∆u + 2∆u + 2t(∆u)t ,

∆v = 2∆u + x · D∆u + 2t(∆u)t .

Hence vt − ∆v = 0, i.e., v satisfies heat equation.

Fourier transform: For any f ∈ L1 (Rn ), we define

i) Fourier transform of f , denoted by fˆ, is defined as
Z
1
fˆ(ξ) = n e−i x·ξ f (x) dx.
(2π) 2 R n

ii) The inverse fourier transform of f , denoted by fˇ, is defined as

Z
1
fˇ(ξ) = n ei x·ξ f (x) dx.
(2π) 2 Rn

We intend to extend the definition for functions f ∈ L2 (Rn ).

Theorem 5.1 (Plancherel’s theorem). Assume that u ∈ L1 ∩ L2 (Rn ). Then û, ǔ ∈ L2 (Rn ) and

kûkL2 (Rn ) = kǔkL2 (Rn ) = kukL2 (Rn ) .

Definition of Fourier transform on L2 : Since L1 ∩ L2 (Rn ) is dense in L2 (Rn ), there exists

a sequence {un } ⊂ L1 ∩ L2 (Rn ) such that

uk → u in L2 (Rn ).

Hence by Plancherel’s theorem

kûk − ûj kL2 (Rn ) = kuk − uj kL2 (Rn ) .

Hence {ûk } is Cauchy sequence in L2 (Rn ). Hence it has a limit, which we defined to be û:

ûk → û in L2 (Rn ).

Theorem 5.2 (Basic properties of Fourier ttransform). Assume that u, v ∈ L2 (Rn ). Then
ˇ ˆ
a) fˆ(−ξ) = fˇ(ξ), fˆ = fˇ = f.
n
b) If u, v ∈ L1 ∩ L2 (Rn ), then f[ ∗ g = (2π) 2 fˆĝ.
c) Ddα f (ξ) = (iξ)α fˆ(ξ) for each multi-index α such that D α u ∈ L2 (Rn ).

2 |ξ|2
Example 5.2. Let f (x) = e−a|x| . Then fˆ(ξ) = 1
n e
− 4a .
(2a) 2
46 A. K. MAJEE

5.1. Derivation of Fundamental solution: Suppose u solves the heat equation

ut − ∆u = 0, in Rn × (0, ∞); u(x, 0) = g(x), x ∈ Rn . (5.1)
Taking the Fourier transform in the x- variable, we get
ût (ξ, t) + |ξ|2 û(ξ, t) = 0, ξ ∈ Rn , t > 0; û(ξ, 0) = ĝ(ξ), ξ ∈ Rn .
This ODE can be solved explicitly
n 2t
2 (2π) 2 ĝ(ξ)e−|ξ| g[∗ F (ξ)
û(ξ, t) = ĝ(ξ)e−|ξ| t = n = n
(2π) 2 (2π) 2
. Consequently, we have
g∗F
u= n
(2π) 2
2
where F̂ (ξ) = e−t|ξ| . Thus, F is given by
 ˇ  1
Z
2
F = e−t|ξ|2 = n ei x·ξ e−t|ξ| dξ.
(2π) 2 Rn
One can easily check that Z
2
π n |x|2
ei x·ξ e−t|ξ| dξ = e−
2
4t .
Rn t
Thus, the solution of the heat equation is given by
Z
1 |x−y|2
u(x, t) = n e− 4t g(y) dy (x ∈ Rn , t > 0).
(4πt) 2 Rn
Definition 5.1. The function Φ defined as
 2
 1 e− |x|4t , x ∈ Rn , t > 0
n
Φ(x, t) := (4πt) 2 (5.2)
0, x ∈ Rn , t < 0

is called the fundamental solution of the heat equation.

Observations:
R
i) Φ is radial in the x-variable and Rn Φ(x, t) dx = 1 for all t > 0.
ii) For each δ > 0, Φ is infinitely differentiable, with uniformly bounded derivatives of all
orders on Rn × [δ, ∞).
iii) Φ ∈ C ∞ (Rn × R \ {(0, 0)}) and solves the heat equation ut − ∆u = 0 in Rn × R \ {(0, 0)}.
Theorem 5.3. Let g ∈ C(Rn ) ∩ L∞ (Rn ). Then the function
Z
u(x, t) = Φ(x − y, t)g(y) dy
Rn
satisfies the followings:
i) u ∈ C ∞ in Rn × (0, ∞),
ii) ut − ∆u = 0 in Rn × (0, ∞),
iii) lim u(x, t) = g(x0 ) for each x0 ∈ Rn .
(x,t)→(x0 ,0), t>0

Proof. Fix t0 > 0. Then Φ and all its derivatives are bounded and integrable in Rn × [t0 , ∞).
Thus, we can pass the derivatives inside the integral. This shows that u ∈ C ∞ in Rn × [t0 , ∞).
Since t0 > 0 is arbitrary, we get u ∈ C ∞ in Rn × (0, ∞). Furthermore, we have for x ∈ R, t > 0
Z Z Z
ut − ∆u = Φt (x − y, t)g(y) dy − ∆x Φ(x − y, t)g(y) dy = (Φt − ∆x Φ)(x − y, t)g(y) dy = 0 .
Rn Rn Rn
 PDES 47

Fix a point x0 ∈ Rn . Let ε > 0 be given. We need to show that there exists a δ > 0 such that
|u(x, t) − g(x0 )| < ε for |(x, t) − (x0 , 0)| < δ.
Sine g is continuous at x0 , there exists δ1 > 0 such that
|g(y) − g(x0 )| < , for y ∈ B(x0 , δ1 ).
We have
Z
|u(x, t) − g(x0 )| = Φ(x − y, t)[g(y) − g(x0 )] dy
n
ZR Z
≤ε Φ(x − y, t) dy + Φ(x − y, t)|g(y) − g(x0 )| dy
B(x0 ,δ1 ) B(x0 ,δ1 )c
Z
≤ ε + 2kgkL∞ (Rn ) Φ(x − y, t) dy ≡ ε + A .
B(x0 ,δ1 )c
δ1
Now, for y ∈ B(x0 , δ1 )c and |x − x0 | < 2, we have
δ1 1
|y − x0 | = |y − x + x − x0 | ≤ |y − x| + |x − x0 | < |y − x| + < |y − x| + |y − x0 |
2 2
=⇒ |y − x0 | ≤ 2|y − x|

Therefore, if y ∈ B(x0 , δ1 )c and |x − x0 | < δ21 , one has |y − x| > δ21 . Hence, we have
Z Z
1 |x−y|2 1 |z|2
− 4t
A≤C n e g(y) dy = C n e− 4t g(y) dy
(4πt) 2 |y−x|> δ21 (4πt) 2 |z|> δ21
Z
1 |y|2
=C n e− 4 g(y) dy → 0 as t → 0+ .
(4π) 2 |y|> 2δ√1t
δ1
Thus, one can choose δ < 2 to conclude the result. 
Remark 5.1. We have the following:
|x−y|2
• Since e− 4t > 0, if g ≥ 0 and g > 0 at some point, then u(x, t) > 0 for all x ∈ Rn
and t > 0. This phenomenon is known as the infinite speed of propagation of
disturbances of the heat equation.
• u(x, t) is bounded. In particular, |u(x, t)| ≤ kgkL∞ (Rn ) .
Example 5.3. Show that there exists a solution u(x, t) of the heat equation
ut − ∆u = 0 in R2 × (0, ∞); u(x, 0) = g(x), x ∈ R2
such that lim u(x, t) = 0, where g ∈ C(R2 ) ∩ L∞ (Rn ) with R2 |g(y)| dy < ∞.
R
t→∞ R
Solution: Let A := R2 |g(y)| dy < +∞. In view of the above theorem
Z
u(x, t) = Φ(x − y, t)g(y) dy
R2

solves the given PDE. We need to check that lim u(x, t) = 0. In other words, given any ε > 0,
t→∞
we need to find M > 0 such that whenever t ≥ M , |u(x, t)| < ε for all x ∈ R2 . Observe that
Z
1 − |x−y|2 1
|u(x, t)| ≤ e 4t |g(y)| dy ≤ A .
R2 4πt 4πt
A
Choose M = 4πε . Then for all t ≥ M , we have |u(x, t)| < ε for all x ∈ R2 .
48 A. K. MAJEE

Example 5.4 (Solution to heat equation in half-line). Let φ be a continuous and bounded
function defined on the domain [0, ∞) with φ(0) = 0. We are interested in finding the solution
of the heat equation
ut − uxx = 0, x > 0, 0 < t < ∞; u(x, 0) = φ(x), x > 0; u(0, t) = 0, t > 0.
We extend φ to an odd function on the whole real line. Define
(
φ(x), x ≥ 0,
φodd (x) := .
−φ(−x), x < 0 .
Then φodd is bounded and continuous function on R. Then the function v(x, t) defined by
Z
v(x, t) := Φ(x − y, t)φodd (y) dy
R
solves the heat equation
vt − vxx = 0 x ∈ R, t > 0; v(x, 0) = φodd (x) x ∈ R.
Restricting the x variable to only the positive half-line produces the function
u(x, t) = v(x, t) .
x≥0
Note that v(x, t) is an odd function of x. Therefore, v(0, t) = 0 = u(0, t). Moreover, u satisfies
the heat equation. Hence the solution of the given heat equation is given by
Z ∞ Z 0
u(x, t) = Φ(x − y, t)φ(y) dy + Φ(x − y, t)φ(−y) dy
0 −∞
Z ∞
 
= Φ(x − y, t) − Φ(x + y, t) φ(y) dy, x > 0, t > 0 .
0

Example 5.5 (Heat equation in half-line with Neumann boundary conditions). Let φ be a
bounded continuous function defined on the domain [0, ∞). We are interested in finding the
solution of the heat equation
ut − uxx = 0, x > 0, 0 < t < ∞; u(x, 0) = φ(x), x > 0; ux (0, t) = 0, t > 0.
Note that derivative of a even function is a odd function and hence the derived function at
origin is equal to zero. With this observation, we extend φ to an even function on the whole real
line. Define (
φ(x), x ≥ 0,
φeven (x) =
φ(−x), x ≤ 0 .
Then v(x, t) defined by
Z Z ∞ 
v(x, t) = Φ(x − y, t)φeven (y) dy = Φ(x − y, t) + Φ(x + y, t) φ(y) dy
R 0
is a solution of the heat equation
vt − vxx = 0 x ∈ R, t > 0; v(x, 0) = φeven (x) x ∈ R.
One can easily check that v(x, t) − v(−x, t)] solves the heat equation with zero initial data, and
therefore, v(x, t) is even function on x. Set u(x, t) := v(x, t) . Then one can easily check
x≥0
that u(x, t) solves the given heat equation. In other words,
Z ∞
 
u(x, t) = Φ(x − y, t) + Φ(x + y, t) φ(y) dy, x > 0, t > 0
0
solves the given heat equation with Neumann boundary conditions.
 PDES 49

5.2. Nonhomogeneous Problem: We would like to study the nonhomogeneous heat equation
with zero initial conditions. Recall that (x, t) 7→ Φ(x − y, t − s) is a solution of heat equation
for given y ∈ Rn , 0 < s < t. For fixed such s, the function
Z
u(x, t; s) = Φ(x − y, t − s)f (y, s) dy
Rn
solves the heat equation
ut (·, ·; s) − ∆u(·, ·; s) = 0 in Rn × (s, ∞); u(·, ·; s) = f (·, s) on Rn × {t = s} . (5.3)
Duhamel’s principle asserts that, we can build a solution of nonhomogeneous equation out
of the solutions of (5.3).
Rt
Theorem 5.4. Let f ∈ Cc2,1 (Rn ×[0, ∞)). Define ū(x, t) := 0 u(x, t; s) ds. Then ū(x, t) satisfies
i) ū ∈ C 2,1 (Rn × (0, ∞)),
ii) ūt − ∆ū = f in Rn × (0, ∞),
iii) lim ū(x, t) = 0 for each y ∈ Rn .
(x,t)→(y,0), t>0

Proof. We first show that

Z tZ
ū(x, t) = Φ(x − y, t − s)f (y, s) dy ds
0 Rn
is in C 2,1 (Rn × (0, ∞)). We note that, since Φ has a singularity at (0, 0), we cannot directly
justify passing the derivatives inside the integral. Using the change of variables ỹ = x − y and
s̃ = t − s, we have
Z tZ
ū(x, t) = Φ(ỹ, s̃)f (x − ỹ, t − s̃) dỹ ds̃ .
0 Rn

Since f ∈ Cc2,1 (Rn × [0, ∞)) and Φ is smooth near s̃ = t > 0, we have
Z tZ Z
ūt (x, t) = Φ(y, s)ft (x − y, t − s) dy ds + Φ(y, t)f (x − y, 0) dy ,
0 Rn Rn
Z tZ
∂ 2 ū ∂2f
(x, t) = Φ(y, s) (x − y, t − s) dy ds (i, j = 1, . . . , n) .
∂xi ∂xj 0 Rn ∂xi ∂xj
Therefore, ū ∈ C 2,1 (Rn × (0, ∞)).
To show ii), we use same change of variables and have
Z tZ Z
ūt (x, t) − ∆x ū(x, t) = Φ(y, s)ft (x − y, t − s) dy ds + Φ(y, t)f (x − y, 0) dy
0 Rn Rn
Z tZ
− Φ(y, s)∆x f (x − y, t − s) dy ds
0 Rn
Z tZ Z
=− Φ(y, s)fs (x − y, t − s) dy ds + Φ(y, t)f (x − y, 0) dy
0 Rn Rn
Z tZ
− Φ(y, s)∆y f (x − y, t − s) dy ds .
0 Rn
We would like to use the integration by parts formula to put the derivatives on Φ and use the
fact that Φ solves the heat equation. However, since Φ has a singularity at t = 0, we break the
integral as follows.
Z tZ
ūt (x, t) − ∆x ū(x, t) = Φ(y, s)[−∂s − ∆y ]f (x − y, t − s) dy ds
ε Rn
50 A. K. MAJEE
Z εZ Z
+ Φ(y, s)[−∂s − ∆y ]f (x − y, t − s) dy ds + Φ(y, t)f (x − y, 0) dy
0 Rn Rn
≡ Iε + Jε + K .
Since f ∈ Cc2,1 (Rn × [0, ∞)) and Rn Φ(y, s) dy = 1 for any s > 0, we see that
R
Z εZ
|Jε | ≤ C Φ(y, s) dy ds ≤ Cε → 0 as ε → 0.
0 Rn
Since f has compact support, by using the integration by parts formula, we have
Z tZ Z
Iε = [∂s − ∆y ]Φ(y, s)f (x − y, t − s) dy ds + Φ(y, ε)f (x − y, t − ε) dy
ε Rn Rn
Z
− Φ(y, t)f (x − y, 0) dy
Rn
Z
= Φ(y, ε)f (x − y, t − ε) dy − K ,
Rn
where in the last equality, we have used the fact that Φ(y, s) solves the heat equation for s > 0.
Thus, we get
Z
ūt (x, t) − ∆x ū(x, t) = lim [Iε + Jε + K] = lim Φ(y, ε)f (x − y, t − ε) dy
ε→0 ε→0 Rn
hZ Z

i
= lim Φ(y, ε)f (x − y, t) dy + Φ(y, ε) f (x − y, t − ε) − f (x − y, t) dy
ε→0 Rn Rn
= lim [A1,ε + A2,ε ] .
ε→0
Note that f is uniformly continuous on the compact support of f . Thus, one has limε→0 A2,ε = 0.
Following the similar calculation as invoked in Theorem 5.3, we arrive at
Z
lim A1,ε = lim Φ(x − y, ε)f (y, t) dy = f (x, t) .
ε→0 ε→0 Rn

In other words, we get

ūt (x, t) − ∆x ū(x, t) = f (x, t).
Observe that
Z tZ
|u(x, t)| ≤ kf kL∞ (Rn ×[0,∞)) Φ(y, s) dy ≤ Ct.
0 Rn
Therefore, as t → 0+ , u(x, t) → 0 as claimed. This completes the proof. 
Regarding non homogeneous heat equation, we arrive at the following theorem:
Theorem 5.5. Let f ∈ Cc2,1 (Rn × [0, ∞)) and g ∈ C(Rn ) ∩ L∞ (Rn ). Then the function u(x, t)
defined by
Z tZ Z
u(x, t) = Φ(x − y, t − s)f (y, s) dy ds + Φ(x − y, t)g(y) dy
0 Rn Rn

is in C 2,1 (Rn × (0, ∞)) ∩ C(Rn × [0, ∞)) and solves the non homogeneous heat equation
ut − ∆u = f in Rn × (0, ∞); u(x, 0) = g(x), x ∈ Rn .
Example 5.6. Write down an explicit formula for a solution of
ut − ∆u + cu = f in Rn × (0, ∞); u = g on Rn × {0}
where c ∈ R is a constant.
 PDES 51

Solution: Let v(x, t) = ect u(x, t). Suppose u solves the given PDE. Then v(x, t) solves the PDE
vt − ∆v = ect f (x, t) in Rn × (0, ∞); v = g on Rn × {0}.
Thus,
Z tZ Z
cs
v(x, t) = Φ(x − y, t − s)e f (y, s) dy ds + Φ(x − y, t)g(y) dy
0 Rn Rn
Z tZ Z
−c(t−s)
=⇒ u(x, t) = Φ(x − y, t − s)e f (y, s) dy ds + Φ(x − y, t)e−ct g(y) dy .
0 Rn Rn
5.3. Mean value formula, maximum principle and uniqueness: Let U ⊂ Rn be an open
and bounded set and T > 0 is fixed. We define the followings:
(i) Parabolic cylinder UT := U × (0, T ).
(ii) Parabolic boundary of UT , denoted as ΓT defined as ΓT := ŪT \ UT : it comprises the
bottom and vertical sides of U × [0, T ], but not the top.
(iii) For fixed x ∈ Rn , t ∈ R, r > 0, we define the heat ball E(x, t; r) as
1
E(x, t; r) := {(y, s) ∈ Rn+1 : s ≤ t, Φ(x − y, t − s) ≥ n } .
r
Let (y, s) ∈ E(x, t; r). Then, we have
n
1 |x−y|2 1 |x−y|2 (4π(t − s)) 2
− 4(t−s) − 4(t−s)
n e ≥ n
=⇒ e ≥
(4π(t − s)) 2 r rn
|x − y|2 r
n r

=⇒ ≤ log p = n log p
4(t − s) 4π(t − s) 4π(t − s)
2 r

=⇒ |x − y| ≤ 4n(t − s) log p .
4π(t − s)
The right hand side of the above inequality is zero if
r2
s = t, or s = t −
.
4π
This implies that (x, t) lies in the boundary of
the heat ball E(x, t; r). Let (y, s) ∈ ∂E(x, t; r).
Then we have |x − y|2 = 4n(t − s) log √ r , and hence
4π(t−s)

n |x − y|2
n ln(r) − ln(4π(t − s)) − = 0.
2 4(t − s)
This shows that the function Ψ : Rn+1 → R defined by
n |x − y|2
Ψ(y, s) := n ln(r) − ln(4π(t − s)) −
2 4(t − s)
vanishes on ∂E(x, t; r).
Theorem 5.6 (Mean value formula). Let u ∈ C 2,1 (UT ) and solves the heat equation ut −∆u = 0.
Then
|x − y|2
Z
1
u(x, t) = n u(y, s) dy ds (5.4)
4r E(x,t;r) (t − s)2
for every E(x, t; r) ⊂ UT .
Proof. Without loss of generality, we assume that (x, t) = (0, 0). Write E(r) := E(0, 0; r).
Define
|y|2 |y|2
ZZ ZZ
1
H(r) := n u(y, s) 2 dy ds = u(ry, r2 s) 2 dy ds.
r E(r) s E(1) s
52 A. K. MAJEE

Differentiating, we have
n
|y|2 2 |y|
2
ZZ X
0
H (r) = yi uyi (ry, r2 s) + 2rus (ry, r s) dy ds
E(1) s2 s
i=1
n
|y|2 |y|2 
ZZ
1 X
= yi uyi (y, s) + 2us (y, s) dy ds
rn+1 E(r) s2 s
i=1
= H1 (r) + H2 (r) .
Note that, the function
n |y|2
Ψ̃(y, s) := n ln(r) − ln(−4πs) +
2 4s
yi
vanishes on ∂E(r), and Ψ̃yi = 2s . Using this, we rewrite H2 (r) as
ZZ n
1 X

H2 (r) = n+1 2us (y, s) 2yi Ψ̃yi (y, s) dy ds
r E(r) i=1
ZZ ZZ
4 4n
= − n+1 y · Ds Dy uΨ̃ dy ds − n+1 us Ψ̃ dy ds (integration by parts w.r.t. y)
r E(r) r E(r)
ZZ ZZ
4 4n
= n+1 y · Dy uDs Ψ̃ dy ds − n+1 us Ψ̃ dy ds (integration by parts w.r.t. s)
r E(r) r E(r)
|y|2 
ZZ ZZ
4  n 4n
= n+1 y · Dy u − − 2 dy ds − n+1 us Ψ̃ dy ds
r E(r) 2s 4s r E(r)
ZZ
1  2n X 
= −A + n+1 − 4nus Ψ̃ − uyi dy ds
r E(r) s y
i
ZZ
1  2n X 
= −A + n+1 − 4n∆y uΨ̃ − uyi dy ds as ut − ∆u = 0
r E(r) s y
i
ZZ
1 2n X 
= −A + n+1 4nDy u · Dy Ψ̃ − uyi dy ds
r E(r) s y
i
n
X yi
= −A (as Dy u · Dy Ψ̃ = uyi ).
2s
i=1

Thus, we have H 0 (r)

= 0, and hence H is constant. In particular,
2 |y|2
2 |y|
ZZ ZZ
H(r) = lim H(t) = lim u(ty, t s) 2 dy ds = u(0, 0) 2
dy ds .
t→0 t→0 E(1) s E(1) s

To prove the mean value property, we need to show that

|y|2
ZZ
2
dy ds = 4.
E(1) s
1
Note that, if (y, s) ∈ E(1), then s ≤ 0 and |y|2 ≤ −2ns log( −4πs ). For the lower bound of s, we
1 1
need to have −4πs ≥ 1 and hence − 4π ≤ s ≤ 0. Thus, we have
Z 0 Z
|y|2
ZZ  ds
2
2
dy ds = |y| dy
E(1) s − 1 |y|2 ≤−2ns log( 1 ) s2
4π −4πs
q
1
Z 0 Z −2ns log( −4πs )  ds
= nα(n)rn−1+2 dr
1
− 4π 0 s2
 PDES 53
Z 0 i n+2 ds
nα(n) h 1 2
= 2n(−s) log( )
n+2 1
− 4π 4π(−s) (−s)2
n+2 Z 1
nα(n)(2n) 2 4π n−2
 1  n+2
2
= s 2 log( ) ds .
n+2 0 4πs
After some calculations, we arrive at
|y|2
ZZ
2
dy ds = 4.
E(1) s
This completes the proof. 
Definition 5.2 (Sub-solution for heat equation). v ∈ C12 (UT ) is called a sub-solution of the
heat equation if vt − ∆v ≤ 0 in UT .
Remark 5.2. Let u ∈ C 2,1 (UT ) be a sub-solution of the heat equation. Then for every
E(x, t; r) ⊂ UT , one has
|x − y|2
Z
1
u(x, t) ≤ n u(y, s) dy ds.
4r E(x,t;r) (t − s)2
Theorem 5.7 (Maximum principle for heat equation). Let u ∈ C 2,1 (UT ) ∩ C(ŪT ) solves the
heat equation in UT . Then
(i) Weak maximum principle: max u = max u.
ŪT ΓT
(ii) Strong maximum principle: in addition if U is connected and there exists a point (x0 , t0 ) ∈
UT such that u(x0 , t0 ) = max u, then u is constant in Ūt0 .
ŪT

Proof. We first show that weak maximum principle for heat equation i.e., (i). Note that if a C 2
function v on UT attains a maximum at any point (x0 , t0 ) ∈ ŪT \ ΓT , then necessarily
vt (x0 , t0 ) − ∆v(x0 , t0 ) ≥ 0.
Assume that u ∈ C 2 (U × (0, T ]). Let ε > 0 be given. Define
v(x, t) := u(x, t) − εt.
Then, v ∈ C 2 (UT ) and
vt − ∆v = ut − ε − ∆u = −ε < 0.
This shows that v cannot attain its maximum anywhere outside the parabolic boundary ΓT . On
the other hand, since ŪT is compact and v ∈ C(ŪT ), v has a maximum in ŪT . Thus, we have
v(x, t) ≤ maxΓT v for any (x, t) ∈ ŪT . Hence, for any (x, t) ∈ ŪT ,
u(x, t) = v(x, t) + εt ≤ max v + εT ≤ max u + εT .
ΓT ΓT
Since ε > 0 is arbitrary, we get
max u = max u.
ŪT ΓT

Now suppose u ∈ ∩ C(ŪT ) solves the heat equation in UT . Then for any T 0 < T ,
C 2,1 (UT )
2
u ∈ C (UT 0 ) and hence, we have
u(x, t) ≤ max u,
ΓT 0
where ΓT 0 is the parabolic boundary of UT 0 . Since ΓT 0 ⊂ ΓT , we get
u(x, t) ≤ max u, for all (x, t) ∈ ŪT 0 .
ΓT

Now for any t < T , we can pick T0

with t < T 0 < T , so the inequality holds. Since u ∈ C(ŪT ),
the inequality holds for t = T . This completes the proof of (i).
54 A. K. MAJEE

Proof of (ii): Let M = maxŪT u, and u(x0 , t0 ) = M for some (x0 , t0 ) ∈ Ū × [0, T ] \ ΓT . Then
for any E(x0 , t0 ; r) ⊂ UT , we have, by mean value property and the fact that
|x0 − y|2
Z
1
1= n dy ds
4r E(x0 ,t0 ;r) (t0 − s)2

|x0 − y|2
Z
1
M = u(x0 , t0 ) = n u(y, s) dy ds ≤ M .
4r E(x0 ,t0 ;r) (t0 − s)2
The equality holds if u is identically equal to M within E(x0 , t0 ; r). Thus,
u(y, s) = M ∀ (y, s) ∈ E(x0 , t0 ; r).
Let (y0 , s0 ) ∈ UT with s0 < t0 be such that the line segment L connecting the points (x0 , t0 )
and (y0 , s0 ) lies in UT .
Claim: u(x, t) = M ∀ (x, t) ∈ L.
Consider
s̃ := min{s ≥ s0 : u(x, t) = M for all points (x, t) ∈ L, s ≤ t ≤ t0 }.
Note that there exists s̄ > 0 such that
{(x, t) ∈ L : t > s̄} ⊂ E(x0 , t0 ; r).
Hence the set {s ≥ s0 : u(x, t) = M for all points (x, t) ∈ L, s ≤ t ≤ t0 } is non-empty. More-
over, since u is continuous, the minimum is attained. If we show that s̃ = s0 , then we are done.
If not, then s̃ > s0 . Then there exists z0 ∈ U with (z0 , s̃) ∈ L ∩ UT such that u(z0 , s̃) = M and
so u ≡ M on E(z0 , s̃; r) for sufficiently small r > 0. Since E(z0 , s̃; r) contains L ∩ {s̃ − σ ≤ t ≤ s̃
for some small σ > 0, it contradicts the fact that s̃ is the minimum. Thus, we conclude that
u(x, t) = M ∀ (x, t) ∈ L.
Finally fix any point (x, t) ∈ UT with 0 ≤ t < t0 . Since U is connected, the exist finite many
points (xi , ti ), 1 ≤ i ≤ m with xm = x and t0 > t1 . . . . > tm = t such that the line segments in
Rn+1 connecting (xi−1 , ti−1 ) to (xi , ti ) lie in UT . Hence u ≡ M on each such segments and so
u(x, t) = M . This completes the proof. 
Remark 5.3. Weak maximum principle holds for a sub-solution of the heat equation. Moreover,
one has the following weak minimum principle. Assume that u ∈ C(ŪT ) ∩ C 2,1 (UT ) satisfies
ut − ∆u ≥ 0. Then
u(x, t) ≥ min u for all (t, x) ∈ ŪT .
ΓT
In other words, u achieves its minimum on the parabolic boundary.
Example 5.7. Show that the function
u(x, t) = 1 − x2 − 2t , 0 ≤ x ≤ 1, 0 ≤ t ≤ T
satisfies the heat equation in UT := (0, 1) × (0, T ). Verify also that maximum principle holds.
Solution: Note that ut = −2 and uxx = −2. Thus ut − ∆u = 0. Moreover, u ∈ C 2,1 (UT ) ∩
C(ŪT ), where U = (0, 1). Observe that for all (x, t) ∈ Ū × [0, T ], 1 − x2 − 2t ≤ 1 and equality
holds if and only if x = 0, t = 0. Hence max u = 1 = u(0, 0). Note here that (0, 0) ∈ ΓT . Again,
ŪT
for all (x, t) ∈ Ū × [0, T ], 1 − x2 − 2t ≥ −2T and equality holds if and only if x = 1, t = T .
Therefore, min u = −2T = u(1, T ).
ŪT

Corollary 5.8. Let u, v ∈ C 2,1 (UT ) and solves the heat equation ut − ∆u = 0 in UT . If u ≤ v
on ΓT , then u ≤ v in UT .
 PDES 55

Example 5.8. Let T > 0 be fixed and U = (0, π). Let u be a solution to the problem
ut − uxx = 0 in UT ; u(0, t) = 0 = u(π, t) ∀t ∈ [0, T ]; u(x, 0) = sin2 (x) ∀x ∈ Ū .
Show that 0 ≤ u(x, t) ≤ e−t sin(x) ∀(x, t) ∈ (0, π) × (0, T ).
Solution: Let v(x, t) = e−t sin x. Then v satisfies the heat equation in UT . Note that v(0, t) =
0 = v(π, t) ∀t ∈ [0, T ]. Moreover, v(x, 0) = sin(x) for all 0 < x < π. Thus, v ≥ 0 on ΓT .
Furthermore, sin2 (x) ≤ sin(x) on [0, π]. Hence u ≤ v on ΓT . Thus, by comparison principle, we
have 0 ≤ u(x, t) ≤ e−t sin(x) ∀(x, t) ∈ (0, π) × (0, T ).
Example 5.9. Suppose u satisfies a heat equation in UT and Φ : R → R is smooth and convex.
Then
a) v = Φ(u) is a subsolution of heat equation in UT .
b) v = |Du|2 + u2t is a sub-solution of heat equation in UT .
Solution: Let v = Φ(u). Then ∆v = Φ00 (u)|∇u|2 + Φ0 (u)∆u. Again, since ut = ∆u, we get that
vt = Φ0 (u)ut = Φ0 (u)∆u. Thus, vt − ∆v = −Φ00 (u)|∇u|2 . Since Φ is smooth and convex, we see
that vt − ∆v ≤ 0i n UT . In other words, v is a subsolution of heat equation in UT .
For part b), let us calculate vt . By using heat equation for u, we notice that
vt = 2Du · Dut + 2ut utt = 2Du · Dut + 2ut (∆u)t .
Moreover, we have
X n Xn n X
X n
2 2
∆v = 2 (uxi xj ) + 2|Dut | + 2Du · D∆u + 2ut (∆u)t = 2 (uxi xj )2 + 2|Dut |2 + vt .
i=1 j=1 i=1 j=1

Thus, vt − ∆v ≤ 0 in UT , and hence v is a subsolution of heat equation in UT .

Based on maximum principle, we arrive at uniqueness of solutions of heat equation.
Theorem 5.9 (Uniqueness on bounded domain). Let g ∈ C(ΓT ), f ∈ C(UT ). Then there exists
at most one solution u ∈ C 2,1 (UT ) ∩ C(ŪT ) of the initial/ boundary value problem
(
ut − ∆u = f in UT
(5.5)
u = g on ΓT .
Example 5.10. Consider the IVP:
ut − uxx = 0 0 < x < 1, t > 0
u(x, 0) = 4x(1 − x), 0≤x≤1
u(0, t) = 0 = u(1, t), t > 0.
Show that u(x, t) = u(1 − x, t) for all 0 ≤ x ≤ 1, t ≥ 0.
Solution: Let v(x, t) = u(1 − x, t). It is easy to check that v satisfies the given PDE. Hence by
uniqueness of solutions, we conclude that v(x, t) = u(x, t). In other words, u(x, t) = u(1 − x, t)
for all 0 ≤ x ≤ 1, t ≥ 0.
Next we show that under certain growth assumption, maximum principle for Cauchy problem
holds.
Theorem 5.10 (Maximum principle for Cauchy problem). Let u ∈ C(Rn × [(0, T ]) ∩ C(Rn ×
[0, T ]) and solves
ut − ∆u = 0 in Rn × (0, T ); u = g on Rn × {t = 0}
and satisfies the growth estimate
2
u(x, t) ≤ Aea|x| (x ∈ Rn , 0 ≤ t ≤ T )
56 A. K. MAJEE

for some constants A, a > 0. Then

sup u = sup g .
Rn ×[0,T ] Rn

As a consequence, we have uniqueness of Cauchy problem under some growth condition.

Corollary 5.11 (Uniqueness of Cauchy problem). Let g ∈ C(Rn ), f ∈ C(Rn × [0, T ]). Then
there exists at most one solution u ∈ C 2,1 (Rn × (0, T ]) ∩ C(Rn × [0, T ]) of the IVP
ut − ∆u = f in Rn × (0, T ); u = g on Rn × {t = 0}
satisfying the growth estimate
2
u(x, t) ≤ Aea|x| (x ∈ Rn , 0 ≤ t ≤ T )
for some constants A, a > 0.
There may be infinitely many solutions of the Cauchy problem for heat equation without the
growth condition.
Example 5.11 (Tychonov’s example). There are infinitely many solutions to the initial value
problem
ut − uxx = 0, x ∈ R, t > 0; u(x, 0) = 0, x ∈ R.
Solution: For some g ∈ C ∞ (R), define
∞
X x2k
u(x, t) = g (k) (t) , x ∈ R, t ∈ R .
(2k)!
k=0

If the series converge in a nice way, we can use term by term differentiation. One can easily
check that
∞ ∞ ∞
X x2k X x2k−2 X x2k
ut = g (k+1) (t) , uxx = g (k) (t) = g (k+1) (t) .
(2k)! (2k − 2)! (2k)!
k=0 k=1 k=0

Thus, u solves the heat equation. Now we choose g ∈ C ∞ (R) as follows: for α > 1
( −α
e−t , t > 0,
gα (t) =
0, t ≤ 0 .
g(t) is real analytic except for t = 0. It remains to verify that
lim u(x, t) = 0 .
t→0+

To do so, we need bound for g (k) (t). One can show that, for some θ ∈ (0, 1)
k! − 1α
|g (k) (t)| ≤ e 2t .
(θt)k
Thus, using above estimate and the fact that (k!)2 ≤ (2k)!
∞ ∞
X k! − 1α |x|2k X 1 |x|2
k − 1α  1 |x|2 
|u(x, t)| ≤ e 2t ≤ e 2t = exp − + .
(θt)k (2k)! k! θt 2tα θt
k=0 k=0

This also shows that the series converge in nice way and the term by term differentiation is
justified. Moreover, u ∈ C ∞ (R × R+ ). Furthermore, on each interval [x1 , x2 ], as t → 0+ ,
u(x, t) → 0 uniformly. Thus, given Cauchy problem has infinitely many solutions.
 PDES 57

5.4. Energy method. We now present an alternate technique to prove uniqueness of solutions
to the heat equation on bounded domains.
Theorem 5.12. Let U be a bounded domain with C 1 -boundary. Then the problem
ut − ∆u = f in U × (0, T ), u = g on ΓT
has at most one solution u ∈ C 2 (UT ) ∩ C 1 (ŪT ).
Proof. Suppose there are two solutions u1 and u2 . Let u = u1 − u2 . Then
ut − ∆u = 0 in U × (0, T ), u = 0 on ΓT .
Define Z
E(t) := u2 (x, t) dx (0 ≤ t ≤ T ).
U
Then E(0) = 0, and
Z Z Z
0
E (t) = 2 uut dx = 2 u∆u dx = −2 |∇u|2 dx ≤ 0 .
U U U
This show that E(t) is non-increasing. Since E(0) = 0 and E(t) ≥ 0, we have E(t) = 0 for all
0 ≤ t ≤ T . Since w ∈ C 1 (ŪT ), we u = 0 in UT . 
Theorem 5.13 (Backward uniqueness). Suppose u1 and u2 are two solutions of the problem
ut − ∆u = f in U × (0, T ), u = g on ∂U × [0, T ] .
If u1 (x, t0 ) = u2 (x, t0 ) for all x ∈ U and for some t0 , then u1 = u2 in U × [0, t0 ].
Proof. Let u = u1 − u2 . Then
ut − ∆u = 0 in U × (0, T ), u = 0 on ∂U × [0, T ], u(x, t0 ) = 0 in U .
Define Z
u2 (x, t) dx (0 ≤ t ≤ T ).
E(t) :=
R UP
Note that E(t0 ) = 0 and E 0 (t) = −2 U ni=1 uxi uxi . Since (ut )xi = (uxi )t , we have, by using
integration by parts formula
Xn Z X n Z n Z
X
00
E (t) = −4 uxi (uxi )t dx = −4 uxi (ut )xi dx = 4 ut uxi xi dx
i=1 U i=1 U i=1 U
Z Z
=4 ut ∆u dx = 4 |∆u|2 dx, (as ut = ∆u) . (5.6)
U U
On the other hand, by using integration by parts formula and Cauchy-Schwartz inequality, we
have
Z Z 1  Z 1
0 2 2 2
E (t) = 2 u∆u dx ≤ 2 u (x, t) dx |∆u(x, t)|2 dx
U U U
Z
0 2
=⇒ [E (t)] ≤ 4E(t) |∆u(x, t)| dx = E(t)E 00 (t) (by (5.6)) .
2
U
If E(t) = 0 for all t ∈ [0, t0 ], then we are done. Suppose E(t) 6= 0 on [0, t0 ]. Then there exist
0 ≤ t1 < t2 ≤ t0 such that
E(t) > 0 ∀ t ∈ [t1 , t2 ), E(t2 ) = 0.
Define the function
f (t) := log(E(t)), t ∈ [t1 , t2 ) .
58 A. K. MAJEE

E(t)E 00 (t)−(E 0 (t))2

Then t 7→ f (t) is convex on [t1 , t2 ) as f 00 (t) = (E(t))2
≥ 0. Consequently, if 0 < λ < 1
and t1 < t < t2 , we have
log(E((1 − λ)t1 + λt)) ≤ (1 − λ) log(E(t1 )) + λ log(E(t))
=⇒ E((1 − λ)t1 + λt) ≤ (E(t1 ))1−λ (E(t))λ .
Sending t → t−
2 and recalling that E(t2 ) = 0, we get

0 ≤ E((1 − λ)t1 + λt2 ) ≤ (E(t1 ))1−λ (E(t2 ))λ = 0, ∀0<λ<1

which contradicts the fact that E(t) > 0 on [t1 , t2 ). This completes the proof. 
5.5. The method of separation of variables for heat equation. We would like to solve
the problem
ut − uxx = 0 in 0 < x < l, t > 0; u(x, 0) = φ(x), x ∈ (0, l); u(0, t) = 0 = u(l, t), t ≥ 0, (5.7)
by using separation of variables, where φ ∈ C[0, l]. Let u(x, t) = X(x)T (t). Then we have
X 00 (x) T 0 (t)
= = −λ
X(x) T (t)
for constant λ. For X, it satisfies two point boundary value problem
X 00 (x) + λX(x) = 0; X(0) = X(l) = 0.
This is a typical Sturm-Liouville type eigenvalue problem. For non-trivial solution, we have seen
2 2
that λn = n l2π , n = 1, 2, . . ., and associated eigen functions are Xn (x) = Bn sin( nπ
l x). We now
substitute the λn in the equation of T , namely Tn (t) + λn Tn (t) = 0 to obtain Tn (t) = an e−λn t .
0

Thus, we formally define

∞
X nπ
u(x, t) = an e−λn t sin( x).
l
n=1
We determine an from the initial condition. Note that u(x, 0) = φ(x) = ∞ nπ
P
n=1 an sin( l x). By
using orthogonal property
Z l (
nπ mπ 0, if m 6= n ,
sin( x) sin( x) dx = l
0 l l 2 , if m = n ,
we see that
2 l
Z
nπ
an = φ(x) sin( x) dx .
l 0 l
P∞
We need to show the convergence of the series n=1 an e−λn t sin( nπ l x). Note that, since φ ∈
C([0, l]), there exists a constant K > 0 such that |an | ≤ K. For any t1 > t0 > 0, we see that
nπ n2 π 2
|an e−λn t sin( x)| ≤ Ke− l2 t on [0, l] × [t0 , t1 ].
l
Therefore, the series converges uniformly and absolutely. One can check continuously differen-
2 2
− n 2π t
tiability of u in t and x as well. Hence u(x, t) = ∞ sin( nπ
P
n=1 na e l
l x) defines the solution
of the given problem.
Example 5.12. Solve the IVP using Fourier method:
ut − uxx = 0, 0 < x < 1, t > 0; u(x, 0) = x, 0 < x < 1; u(0, t) = 0 = u(1, t), t > 0 .
Solution: The solution is given by
∞ Z 1
−n2 π 2 t
X
u(x, t) = fn e sin(nπx), where fn = 2 x sin(nπx) dx.
n=1 0
 PDES 59

2
One can use integration by parts formula to obtain fn = −(−1)n nπ . Hence the solution is given
∞
2 2
X
by u(x, t) = − (−1)n e−n π t sin(nπx).
n=1

Remark 5.4. One can use separation of variables method to find solution of heat equation with
periodic and Neumann boundary conditions as follows.
• Separation of variable method for periodic boundary conditions: One can easily
deduce that all the solution of the ODE
−y 00 = λy, x ∈ (−l, l); y(−l) = y(l), y 0 (−l) = y 0 (l)
are given by
nπ 2 nπ nπ
λ0 = 0, y0 (x) = c0 6= 0; λn = ( ) , yn (x) = cn cos( x) + Dn sin( x), n = 1, 2, 3, . . .
l l l
0
Moreover, the solutions for Tn satisfying Tn (t) + λn Tn (t) = 0 are given by
(
An e−λn t , n = 1, 2, . . .
Tn (t) = An e−λn t =
A0 , n = 0
Therefore, the solution of the heat equation
ut − uxx = 0, x ∈ (−l, l), t > 0; u(x, 0) = φ(x), x ∈ (−l, l)
u(−l, t) = u(l, t), ux (−l, t) = ux (l, t), t ≥ 0
is given by the formula
∞
X nπ nπ  nπ 2
An cos( x) + Bn sin( x) e−( l ) t

u(x, t) = A0 +
l l
n=1
where the coefficients A0 , An and Bn are given by
1 l 1 l 1 l
Z Z Z
nπ nπ
A0 = φ(x) dx, An = cos( x)φ(x) dx, Bn = sin( x)φ(x) dx .
2l −l l −l l l −l l
• Separation of variable method for Neumann boundary conditions: Consider
the following heat equation with Neumann boundary conditions
ut − c2 uxx = 0, 0 < x < L, t > 0; u(x, 0) = φ(x), 0 < x < L,
ux (0, t) = ux (L, t) = 0 t > 0 .
Then the solution of the underlying heat equation is given by the formula
∞
2
X
u(x, t) = a0 + an cos(µn x)e−λn t ,
n=1
where
Z L Z L
nπ 1 2
µn = , λn = cµn , a0 = φ(x) dx, an = φ(x) cos(µn x) dx .
L L 0 L 0
Example 5.13. Solve the heat equation
ut − uxx = 0, x ∈ (−1, 1), t > 0; u(x, 0) = sin(πx) + cos(πx), x ∈ (−1, 1),
u(−1, t) = u(1, t), ux (−1, t) = ux (1, t) t ≥ 0.
Solution: Using separation of variables, the solution of the given heat equation is given by
∞
2 2
X
An cos(nπx) + Bn sin(nπx) e−n π t ,
 
u(x, t) = A0 +
n=1
60 A. K. MAJEE

where the coefficients A0 , An and Bn are given by

1 1
Z Z 1 Z 1
A0 = (sin(πx) + cos(πx)) dx, An = cos(nπx)φ(x) dx, Bn = sin(nπx)φ(x) dx .
2 −1 −1 −1
One can easily check that A0 = 0, An = Bn = 0 for n ≥ 2 and A1 = 1 = B1 . Thus,
2
u(x, t) = cos(πx) + sin(πx) e−π t .

Example 5.14. Solve the following problem

1
ut − uxx = 0, 0 < x < 1, t > 0
4
ux (0, t) = ux (1, t) = 0, t > 0
u(x, 0) = 100 x(1 − x), 0 < x < 1 .
Solution: The solution is given by
∞
X n2 π 2
u(x, t) = a0 + an cos(nπx)e− 4
t
,
n=1
where the coefficients are determined by the formula
Z 1 Z 1
50 200
100x(1 − x) cos(nπx) dx = − 2 2 1 + (−1)n , n ≥ 1 .


a0 = 100x(1 − x) dx = , an = 2
0 3 0 n π
Thus, the solution is
∞
50 100 X 1 −k2 π2 t
u(x, t) = − 2 e cos(2kπx).
3 π k2
k=1

5.5.1. Inhomogeneous heat equation on bounded domains. Consider the initial/boundary

value problem for the inhomogeneous heat equation on an interval I ⊂ R

2
ut − k uxx = f (x, t), x ∈ I, t > 0

u(x, 0) = φ(x), x ∈ I

u satisfies certain BCs, t > 0 .


Using Duhamel’s principle, we expect the solution to be given by

Z t
u(x, t) = S(t)φ(x) + S(t − s)f (x, s) ds,
0
where S(t) is the solution operator associated with the homogeneous problem. For example,
solution to the associated homogeneous problem, in case of symmetric boundary conditions, is
given by
∞
2
X
uh (x, t) = An e−k λn t Xn (t)
n=1
where Xn are the eigenfunctions and λn the corresponding eigenvalues of the eigenvalue problem
−X 00 = λX, x ∈ I, X satisfies symmetric boundary conditions
and the coefficients An are defined by
hXn , φi
An = .
hXn , Xn i
Therefore, the solution operator associated with the homogeneous equation is given by
∞
2
X
S(t)φ(x) = An e−k λn t Xn (x).
n=1
 PDES 61

Hence the solution of the given inhomogeneous problem is

∞ ∞
Z tX
2 2
X
u(x, t) = An e−k λn t Xn (x) + Bn (s)e−k λn (t−s) Xn (x) ds,
n=1 0 n=1

where Bn (s) is determined by the formula

hXn , f (·, s)i
Bn (s) = .
hXn , Xn i
Example 5.15. Solve the inhomogeneous heat equation:
ut − uxx = e−t sin(πx), 0 < x < 1, t > 0
u(0, t) = u(1, t) = 0, t>0
u(x, 0) = sin(2πx), 0 < x < 1.
Solution: Using the method of separation of variables, the solution of the associated homoge-
neous equation is
∞
2 2
X
uh (x, t) = an e−n π t sin(nπx) ,
n=1
where R1 (
0sin(2πx) sin(nπx) dx 1, n = 2
an = R1 2 =
0 sin (nπx) dx
0, n 6= 2 .
Therefore, the solution to the inhomogeneous equation is given by
∞
Z tX
−4π 2 t 2 2
u(x, t) = e sin(2πx) + bn (s)e−n π (t−s) sin(nπx) ,
0 n=1

where the function bn (s) is given by

R 1 −s (
0 e sin(πx) sin(nπx) dx e−s , n = 1,
bn (s) = R1 2 =
0 sin (nπx) dx
0, n 6= 1 .
Simplifying the calculation, we have
Z t
2 2t 2
u(x, t) = e−4π t sin(2πx) + sin(πx)e−π e−s+π s ds .
0
5.5.2. Heat equation with inhomogeneous boundary data. Consider the heat equation
with inhomogeneous boundary data
ut − k 2 uxx = 0, 0 < x < l, t > 0; u(x, 0) = φ(x), 0<x<l
u(0, t) = g(t), u(l, t) = h(t), t > 0,
for some differentiable functions g, h on (0, ∞). Introduce a new function
1 
U (x, t) = (l − x)g(t) + xh(t) , 0 < x < l, t > 0 .
l
Define
v(x, t) = u(x, t) − U (x, t).
One can easily check that v satisfies the following inhomogeneous heat equation with homoge-
neous boundary condition

2
vt − k vxx = −U
 U t (x, t), 0 < x < l, t > 0,
v(0, t) = 0 = v(l, t), t > 0,

v(x, 0) = φ(x) − U (x, 0) .

62 A. K. MAJEE

Once we solve for v by using the technique of previous subsection, we can find the solution of
the original problem as
u(x, t) = v(x, t) + U (x, t).
Example 5.16. Solve the heat equation:
ut − uxx = 0, 0 < x < 1, t > 0,
u(x, 0) = 1, 0 < x < 1,
u(0, t) = 1, u(l, t) = 0, t > 0.
Solution. Here U (x, t) = 1 − x, and v(x, t) satisfies the heat equation
vt − uxx = 0, 0 < x < 1, t > 0,
v(x, 0) = x, 0 < x < 1,
v(0, t) = 0 = v(1, t), t > 0.
We have seen that solution to the above problem is given by
∞
2 2
X
v(x, t) = − (−1)n e−n π t sin(nπx) .
n=1
Thus, the solution u(x, t) of the given heat equation is
∞
2 2
X
u(x, t) = 1 − x − (−1)n e−n π t sin(nπx) .
n=1
 PDES 63

6. Wave equation
We will study linear hyperbolic equation namely wave equation utt − ∆u = 0.

6.1. Cauchy Problem for one dim. wave equation: Consider the Cauchy problem
utt − uxx = 0, x ∈ R, t > 0,
u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ R , (6.1)
where g and h are given functions. The equation can be written as
∂ ∂
∂ ∂

+ − u = 0.
∂t ∂x ∂t ∂x
∂u ∂u
Set v(x, t) = ∂t − ∂x . Then vt + vx = 0. This is a transport equation. Thus, solution is given
by
v(x, t) = r(x − t), where v(x, 0) = r(x).
Substituting this value of v, we have
ut − ux = r(x − t), x ∈ R, t > 0; u(x, 0) = g(x).
This is a nonhomogeneous transport equation. Hence the solution is given by
Z t
1 x+t
Z
u(x, t) = g(x + t) + r(x + t − 2s) ds = g(x + t) + r(y) dy .
0 2 x−t
Observe that
r(x) = v(x, 0) = ut (x, 0) − ux (x, 0) = h(x) − g 0 (x) .
Thus, we have, for any x ∈ R, t ≥ 0
1 x+t
Z
h(y) − g 0 (y) dy


u(x, t) = g(x + t) +
2 x−t
 1 x+t
Z
1
= g(x − t) + g(x + t) + h(y) dy . (6.2)
2 2 x−t
Formula (6.2) is called d’Alembert’s formula.
Theorem 6.1. If g ∈ C 2 (R) and h ∈ C 1 (R), then u defined by d’Alembert’s formula solves the
Cauchy problem (6.1).
Remark 6.1. We have the following:
i) The above formula is of the form u(x, t) = F (x + t) + G(x − t) for some F and G.
Conversely, let u(x, t) = F (x + t) + G(x − t) for some F and G. Then utt − uxx = 0.
Note that a general solution of ut − ux = 0 is of the form F (x + t) and general solution
of ut + ux = 0 is of the form G(x − t). Therefore, a general solution of utt − uxx = 0 is
the sum of general solution of ut − ux = 0 and ut + ux = 0.
ii) Solution depends only on (x − t, x + t), which is different from the heat equation.
iii) u(x, t) = 0 for all |x| > R + t if g(x) and h(x) vanish for |x| > R for some R > 0.
Uniqueness: Suppose the problem (6.1) has two solutions, say u1 and u2 . Take v = u1 − u2 .
Then u satisfies the wave equation (6.1) with g = 0 = h. Note that v is C 2 with v(x, t) = 0 for
|x| > 1 + t. Define the energy
Z
1
E(t) = (v 2 + vx2 ) dx .
2 R t
64 A. K. MAJEE

By using integration by parts formula along with differentiation theorem under the integral sign
(which is possible as v has compact support and C 2 - function), we have
Z Z Z
0
E (t) = (vt vtt + vx vxt ) dx = (vt vxx + vx vxt ) dx = (vx vt )x dx = 0 .
R R R
Note that E(0) = 0 and hence E(t) = 0 for all t. Since vt and vx are continuous functions, we
see that v(x, t) is constant. Since v(0, 0) = 0, we get v(x, t) = 0 for all x ∈ R and t ≥ 0. In other
words, Cauchy problem (6.1) has unique solution.
Example 6.1. Solve the IVP:
utt − uxx = 0, x ∈ R, t > 0
u(x, 0) = x3 , ut (x, 0) = sin(x), x ∈ R.
Solution: By d’Alembert’s formula, the solution of the given IVP is given by
 1 x+t
Z
1 1
u(x, t) = (x + t)3 + (x − t)3 + sin(y) dy = x3 + 3xt2 + [cos(x − t) − cos(x + t)]
2 2 x−t 2
= x3 + 3xt2 + sin(t) sin(x) .
Example 6.2. Let u ∈ C 2 (R × [0, ∞)) solves the IVP:
utt − uxx = 0 in R × (0, ∞),
u = g, ut = h on R × {t = 0}.
Suppose g, h have compact support. The kinetic energy resp. potential energy is
Z Z
1 2 1
K(t) := u (x, t) dx, resp. P (t) := u2 (x, t) dx.
2 R t 2 R x
Prove that
a) K(t) + P (t) is constant in t.
b) For large enough times t, K(t) = P (t).
Solution: Since u solves the wave equation, we have
Z Z Z
d
K(t) = ut (x, t)utt (x, t) dx = ut (x, t)uxx (x, t) dx = − utx (x, t)ux dx + boundary term
dt R R R
Note that the solution is given by d’Alembert’s formula
1 x+t
Z
1
u(x, t) = [g(x + t) + g(x − t)] + h(y) dy.
2 2 x−t
Since g, h have compact support, we see that the boundary term vanishes. Hence
Z Z
d d 1  d
K(t) = − utx (x, t)ux dx = − u2x (x, t) dx = − P (t).
dt R dt 2 R dt
In other words, K(t) + P (t) is constant in t.
Since g and h have compact support, there exists a positive constant M such that supp g0 ⊂
[−M, M] and supp h ⊂ [−M, M]. Fix t > M . Then the followings hold:
g(x + t) = h(x + t) = 0 for x > 0 ,
g(x − t) = h(x − t) = 0 for x < 0 ,
g 0 (t) = g 0 (−t) = h(t) = h(−t) = 0 .
Hence by using d’Alembert’s formula , we have for t > M and x > 0,
1 2 1 1
u2t (x, t) = g 0 (x − t) + h2 (x − t) − g 0 (x − t)h(x − t) = u2x (x, t).
4 4 2
 PDES 65

Similarly, we have u2t (x, t) = u2x (x, t) for all t > M with x ≤ 0. Hence K(t) = P (t) for large
enough time t.
6.2. Non-homogeneous problem: For non-homogeneous problem, one can use Duhumel’s
principle to find the solution. For example, consider the one dimensional non-homogeneous
wave equation
(
utt − c2 uxx = f (x, t) in R × (0, ∞)
u(x, 0) = g(x), ut (x, 0) = h(x) on R .
If f ∈ C 1,0 (R × (0, ∞)), g ∈ C 2 (R) and h ∈ C 1 (R), then solution of the above non-homogeneous
problem is given by
1 x+ct 1 t  x+c(t−s)
Z Z Z
1  
u(x, t) = g(x + ct) + g(x − ct) + h(y) dy + f (y, s) dy ds .
2 2c x−ct 2c 0 x−c(t−s)

Example 6.3. Solve the IVP:

utt − 4uxx = 2t, u(x, 0) = x2 , ut (x, 0) = 1.
Solution: This is a non-homogeneous problem. We use Duhumel’s principle to find explicit
formula for the solution. The solution is given by
 1 x+2t 1 t  x+2(t−s)
Z Z Z
1 2 2

u(x, t) = (x + 2t) + (x − 2t) + dy + 2y dy ds
2 4 x−2t 4 0 x−2(t−s)
Z t
1
= x2 + 4t2 + t + {(x + 2(t − s))2 − (x − 2(t − s))2 } ds
4 0
1 4

= x2 + 4t2 + t + x2 t + t3 .
2 3
Example 6.4. Solve the IVP:
utt − uxx = tex , x ∈ R, t > 0; u(x, 0) = 0 = ut (x, 0), x ∈ R.
Solution: By Duhumel’s principle, the solution of the given non-homogeneous wave equation is
given by
!
1 t
Z Z x+(t−s)
u(x, t) = sey dy ds
2 0 x−(t−s)
Z t
1 1 x+t
s ex+t−s − ex+s−t ds = − ex−t − 2tex .



= e
2 0 2
6.3. Reflection method: Consider the initial/boundary-value problem on the half space x > 0:
utt − uxx = 0, in R+ × (0, ∞),
u = g, ut = h, on R+ × {t = 0} ,
u = 0 on {x = 0} × (0, ∞) , (6.3)
where g, h are given with g(0) = 0 = h(0). Set
( ( (
u(x, t), x ≥ 0, t ≥ 0 g(x), x ≥ 0 h(x), x ≥ 0
ũ(x, t) = ; g̃(x) = ; h̃(x) = .
−u(−x, t) x ≤ 0, t ≥ 0 −g(−x), x ≤ 0 −h(−x), x ≤ 0
Then the problem (6.3) becomes

ũtt − uxx = 0, in R × (0, ∞),

66 A. K. MAJEE

ũ = g̃, ũt = h̃, on R × {t = 0} .

Hence, by d’Alembert’s formula
Z x+t
1  1
ũ(x, t) = g̃(x − t) + g̃(x + t) + h̃(y) dy
2 2 x−t

Note that for x−t > 0, g̃(x±t) = g(x±t) and h̃(x±t) = h(x±t). For x−t < 0, g̃(x−t) = −g(t−x).
Hence, the solution of (6.3) is given by for any x ≥ 0, t ≥ 0:
 1 x+t
 Z
1
g(x − t) + g(x + t) + h(y) dy, if x ≥ t


2 2 x−t

u(x, t) =  1 x+t
Z
1
 − g(t − x) + g(x + t) + h(y) dy, if 0 ≤ x ≤ t .


2 2 t−x
Example 6.5. Consider the wave equation

utt − uxx = 0, x > 0, t > 0,

u(x, 0) = 0 = ut (x, 0), 0 ≤ x < ∞,

u(0, t) = 1 + t, t ≥ 0 .


Define v(x, t) = u(x, t) − (1 + t). Then v solves the PDE


vtt − vxx = 0, x > 0, t > 0,

v(x, 0) = 0, vt (x, 0) = −1, 0 ≤ x < ∞,

v(0, t) = 0, t ≥ 0 .


Hence by reflection method, solution v is given by

(
−t, if x ≥ t
v(x, t) =
−x, if 0 ≤ x ≤ t .
Hence the solution of the given problem is given by
(
1, if x ≥ t
u(x, t) =
1 + t − x, if 0 ≤ x ≤ t .
Remark 6.2. In the case of Neumann boundary condition i.e., ux = 0 on {x = 0} × (0, ∞) in
(6.3), one can use even reflection to get the solution as
 1 x+t
 Z
1
 g(x − t) + g(x + t) + h(y) dy, if x ≥ t


2 2 x−t
u(x, t) =  1  x+t
Z Z t−x
1 
 g(t − x) + g(x + t) + h(y) dy , if 0 ≤ x ≤ t .

 h(y) dy +
2 2 0 0
6.4. Cauchy problem for higher dimension: Consider the Cauchy problem

utt − uxx = 0, x ∈ Rn , t > 0,

u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ Rn . (6.4)
Suppose u solves the above Cauchy problem. Define for x ∈ Rn , t > 0 and r > 0
Z Z
1 1
u(x, t; r) = u(y, t) dσ(y); g(x; r) = g(y) dσ(y);
σ(∂B(x, r)) ∂B(x,r) σ(∂B(x, r)) ∂B(x,r)
Z
1
h(x; r) = h(y) dσ(y) .
σ(∂B(x, r)) ∂B(x,r)
 PDES 67

Fix x ∈ Rn . For (r, t) ∈ R+ × (0, ∞), define v(r, t) = u(x, t; r). Following the similar
calculation as done in Mean-value theorem for harmonic function, we see that
Z Z
r 1 1
vr (r, t) = ∆u(y, t) dy = n−1 ∆u(y, t) dy
n |B(x, r)| B(x,r) nr αn B(x,r)
Z Z
n−1 1 1
=⇒ r vr = ∆u(y, t) dy = utt (y, t) dy
nαn B(x,r) nαn B(x,r)
Z
1
=⇒ rn−1 vr r =


utt (y, t) dσ(y)
nαn ∂B(x,r)
n−1
Z
1
=⇒ vrr + vr = n−1 utt (y, t) dσ(y)
r r nαn ∂B(x,r)
Z
1
= utt (y, t) dσ(y) = vtt (r, t) .
σ(∂B(x, r)) ∂B(x,r)
Thus, v solves the following PDE
n−1
vtt = vrr + vr in R+ × (0, ∞); v(r, 0) = g(x; r), vt (r, 0) = h(x, r) . (6.5)
r
The PDE (6.5) is called Euler-Poisson-Darboux equation.
6.4.1. The case n = 3: Kirchhoff ’s formula. Define
ṽ(r, t) = rv(r, t), g̃(r) = rg(x; r), h̃(r) = rh(x; r).
Then ṽ satisfies
ṽtt − ṽrr = 0, in R+ × (0, ∞),
ṽ = g̃, ṽt = h̃, on R+ × {t = 0} ,
ṽ = 0 on {r = 0} × (0, ∞) . (6.6)
Indeed, by using Euler-Poisson-Darboux equation, we have
 2 
ṽtt = rvtt = r vrr + vr = rvrr + 2vr = (v + rvr )r = ṽrr .
r
Hence, we find for 0 ≤ r ≤ t
 1 t+r
Z
1
ṽ(r, t) = g̃(r + t) − g̃(t − r) + h̃(y) dy . (6.7)
2 2 t−r
Note that
h g̃(r + t) − g̃(t − r) Z t+r
ṽ(r, t) 1 i
u(x, t) = lim u(x, t; r) = lim v(r, t) = lim = lim + h̃(y) dy
r→0 r→0 r→0 r r→0 2r 2r t−r
∂
= g̃ 0 (t) + h̃(t) =


tg(x; t) + th(x; t)
Z ∂t Z
1 ∂ 1  ∂
= h(y) dσ(y) + g(y) dσ(y) = t g(x; t) + g(x; t) + th(x; t) .
4πt ∂B(x,t) ∂t 4πt ∂B(x,t) ∂t
One can check that
y − x

Z
∂ 1
g(x; t) = Dg(y) · dσ(y).
∂t σ(∂B(x, t)) ∂B(x,t) t
Therefore, we conclude that
Z
1
u(x, t) = th(y) + g(y) + Dg(y) · (y − x) dσ(y) . (6.8)
σ(∂B(x, t)) ∂B(x,t)
The formula (6.8) is called Kirchhoff ’s formula.
68 A. K. MAJEE

Theorem 6.2. Let g ∈ C 3 (R3 ) and h ∈ C 2 (R3 ). Then u defined by the Kirchhoff ’s formula
is in C 2 (R3 × (0, ∞)) and solves the Cauchy problem (6.4) with n = 3.
Example 6.6. Find the solution of (6.4) with n = 3 for g ≡ 0 and h,the characteristic function
on closed unit ball.
Solution: The solution is given by the formula
Z
1
u(x, t) = h(y) dσ(y).
4πt ∂B(x,t)
If the sphere ∂B(x, t) lies completely outside the unit ball i.e., if t > 1 + |x| or t < |x| − 1,
then u(x, t) = 0. If the sphere ∂B(x, t) lies completely inside the unit ball i.e., t < 1 − |x|,
then u(x, t) = t. If the sphere ∂B(x, t) lies partially in the unit ball, i.e., for t such that
|x| − 1 < t < |x| + 1 or t > 1 − |x|,
then we find the surface area of the part of the sphere
πt
∂B(x, t) ∩ B(0, 1) as |x| 1 − (t − |x|)2 . Hence the solution is given by

t, 0 ≤ t 2≤ 1 − |x| ,

u(x, t) = 1−(t−|x|)
4|x| , |x| − 1 ≤ t ≤ 1 + |x| ,

0, 0 ≤ t ≤ |x| − 1 or t > 1 + |x| .


Example 6.7. Let u solves the following wave equation

utt − ∆u = 0 in R3 × (0, ∞), u = 0, ut = h on R3 × {t = 0}
where h is a smooth function and has compact support. Show that there exists a constant C such
that
C
|u(x, t)| ≤ , x ∈ R3 , t > 0.
t
Solution: From the Kirchoff ’s formula, solution u is given by
Z
1
u(x, t) = h(y) dσ(y).
4πt ∂B(x,t)
Suppose the function h supported in the ball B(0, R) for some R > 0. Let A be the surface
area of ∂B(x, t) ∩ B(0, R). Then the maximum value of A will be when the sphere ∂B(x, t) is
completely inside the ball B(0, R). Hence A ≤ 4πR2 . Thus, we have
A R2 C
|u(x, t)| ≤ sup |h(y)| ≤ sup |h(y)| =
4πt y∈R3 t y∈R3 t
where C = R2 sup |h(y)| > 0 and independent of x.
y∈R3

6.4.2. The case n = 2: Poisson formula. Let us write ū(x1 , x2 , x3 , t) = u(x1 , x2 , t). Then
equation (6.4) implies that
(
ūtt − ∆ū = 0 in R3 × (0, ∞)
(6.9)
ū = ḡ, ūt = h̄ on R3 × {t = 0}
where ḡ(x1 , x2 , x3 ) = g(x1 , x2 ) and h̄(x1 , x2 , x3 ) = h(x1 , x2 ). For x = (x1 , x2 ) ∈ R2 , we define
x̄ = (x, 0) ∈ R3 . Then by Kirchhoff’s formula, we have
Z Z
d 1  1
u(x, t) = ū(x̄, t) = t ḡ(y) dσ(y) + t h̄(y) dσ(y) .
dt σ(∂B(x̄, t)) ∂B(x̄,t) σ(∂B(x̄, t)) ∂B(x̄,t)
Observe that Z Z
ḡ(y) dσ(y) = 2 ḡ(y) dσ(y).
∂B(x̄,t) ∂B(x̄,t)∩{y3 >0}
 PDES 69

1
Note that ∂B(x̄, t) ∩ {y3 > 0} is the graph of the function f (y) = t2 − |y − x|2 2
for y ∈
B(x, t) ⊂ R2 . Therefore,
Z Z

1
2 ḡ(y) dσ(y) = 2 g(y) 1 + |∇f |2 2 dy .
∂B(x̄,t)∩{y3 >0} B(x,t)
Observe that
x−y |x − y|2 1 t
∇f = p =⇒ |∇f |2 = 2 =⇒ 1 + |∇f |2 2
=p .
t2 − |y − x|2 t − |y − x|2 t − |x − y|2
2

Thus, we have
Z Z
dh 1 g(y) 1
i h(y)
u(x, t) = p dy + p dy .
dt 2π B(x,t) t2 − |x − y|2 2π B(x,t) t2 − |x − y|2
This can be re-written as
tg(y) + t∇g(y) · (y − x) + t2 h(y)
Z
1
u(x, t) = p dy . (6.10)
2|B(x, t)| B(x,t) t2 − |x − y|2
The formula (6.10) is known as Poisson’s formula.
Theorem 6.3. Let g ∈ C 3 (R2 ) and h ∈ C 2 (R2 ). Then u defined by the Poisson formula (6.10)
is in C 2 (R2 × (0, ∞)) and solves the Cauchy problem (6.4) with n = 2.
6.5. Separation of variables method for wave equation. Consider the equation
utt − uxx = 0, 0 < x < l, t > 0,
u(x, 0) = g(x), ut (x, 0) = h(x), 0 < x < l,
u(0, t) = 0 = u(l, t), t ≥ 0 .
We would like to solve the above problem by using separation of variables for given g and h with
appropriate regularity. Let u(x, t) = X(x)T (t) solves the PDE. Then one has

X 00 (x) T 00 (t)
= = −λ (say).
X(x) T (t)
For X, it satisfies the boundary value problem
X 00 (x) + λX(x) = 0; X(0) = 0 = X(l).
As seen before, this problem has a countable number of solutions
p n2 π 2
Xn (x) = sin( λn x), with λn = 2 .
l
00 (t) √ √
Observe that TT (t) = −λ has two linearly independent solutions sin( λt) and cos( λt). For
λ = λn , we formally define
∞ 
X nπ nπ  nπ
u(x, t) = an sin( t) + bn cos( t) sin( x) . (6.11)
l l l
n=1
Since u(x, 0) = g(x) and ut (x, 0) = h(x), one can determine the coefficients an and bn which are
given by
Z l
2 l
Z
2 nπ nπ
an = h(x) sin( x) dx, bn = g(x) sin( x) dx .
nπ 0 l l 0 l
One can check that the series in the right hand side of (6.11) converges uniformly and absolutely.
Moreover, it is two times differentiable in both the variables. Hence u(x, t) defined in (6.11) is
the unique solution of the given problem.
70 A. K. MAJEE

Example 6.8. Solve the IVP:

utt − uxx = 0, 0 < x < l, t > 0,
u(x, 0) = x(1 − x), ut (x, 0) = x, 0 < x < l,
u(0, t) = 0 = u(l, t), t ≥ 0 .
Solution: We use Fourier method to have explicit form of the solution u(x, t) as
∞ 
X nπ nπ  nπ
u(x, t) = an sin( t) + bn cos( t) sin( x)
l l l
n=1
where an and bn are given by
Z l
2 l
Z
2 nπ nπ
an = x sin( x) dx, bn = x(1 − x) sin( x) dx .
nπ 0 l l 0 l
It is easy to check that
2 2
an = −2(−1)n ; bn = (−1)n 1 + l(1 − l) .


−
nπ nπ
Remark 6.3. One can use separation of variables (Fourier method) to solve the problem
utt − c2 uxx = 0, 0 < x < l, t > 0,
u(x, 0) = g(x), ut (x, 0) = h(x), 0 < x < l,
u(0, t) = 0 = u(l, t), t ≥ 0 .
The unique solution of the above problem is given by
∞ 
X cnπ cnπ  nπ
u(x, t) = an sin( t) + bn cos( t) sin( x),
l l l
n=1
where the coefficients an and bn are determined by the relation
Z l
2 l
Z
2 nπ nπ
an = h(x) sin( x) dx, bn = g(x) sin( x) dx .
cnπ 0 l l 0 l
Example 6.9. Solve the IVP
utt − 4uxx = 0, 0 < x < 1, t > 0,
u(x, 0) = sin(5πx) + 2 sin(7πx), ut (x, 0) = 0, 0 < x < l,
u(0, t) = 0 = u(1, t), t ≥ 0 .
Solution: Solution of the above problem is given by
∞
X
u(x, t) = bn cos(2nπt) sin(nπx),
n=1
where bn is given by the formula
Z 1


bn = 2 sin(5πx) + 2 sin(7πx) sin(nπx) dx
0
1 if n = 5,

= 2 if n = 7,

0, otherwise .


Thus, the solution is

u(x, t) = cos(10πt) sin(5πx) + 2 cos(14πt) sin(7πx) .
 PDES 71

Remark 6.4. Consider the wave equation with Neumann boundary conditions:

2
utt − c uxx = 0, 0 < x < L, t > 0,

ux (0, t) = 0 = ux (L, t), t > 0,

u(x, 0) = f (x), ut (x, 0) = g(x), 0 ≤ x ≤ L .


Suppose f and g are continuous and piecewise smooth functions on [0, L]. Then, one can use
separation of variables method to get the solution. Moreover, the solution u(x, t) is given by
X cnπ L cnπ

nπ
u(x, t) = α0 + β0 t + αn cos( t) + βn sin( t) cos( x),
L cnπ L L
n≥1

where the coefficients αn and βn are given by

1 L 2 L
Z Z
nπ
α0 = f (x) dx, αn = f (x) cos( x) dx, (n ≥ 1),
L 0 L 0 L
Z L Z L
1 2 nπ
β0 = g(x) dx, βn = g(x) cos( x) dx, (n ≥ 1) .
L 0 L 0 L
Example 6.10. Solve the wave equation


 utt − 9uxx = 0, 0 < x < 1, t > 0,

u (0, t) = 0 = u (1, t), t > 0,
x x
u(x, 0) = cos(πx) + 3 cos(5πx), 0 ≤ x ≤ 1,


ut (x, 0) = 1 + cos(5πx) − 9 cos(9πx), 0 ≤ x ≤ 1 .


Solution: Using separation of variables, the solution of the given wave equation is
X βn

u(x, t) = α0 + β0 t + αn cos(3nπLt) + sin(3nπt) cos(nπx),
3nπ
n≥1

where the coefficients αn and βn are given by

Z 1
α0 = (cos(πx) + 3 cos(5πx)) dx = 0,
0

Z L 1, if n = 1

αn = 2 (cos(πx) + 3 cos(5πx)) cos(nπx) dx = 3, if n = 5
0 
0, otherwise,

Z 1
β0 = (1 + cos(5πx) − 9 cos(9πx)) dx = 1,
0

1 Z 1, if n = 5

βn = 2 (1 + cos(5πx) − 9 cos(9πx)) cos(nπx) dx = −9, if n = 9
0 
0, otherwise.


Therefore, solution is given by

1
u(x, t) = t + cos(3πt) cos(πx) + 3 cos(15πt) cos(5πx) + sin(15πt) cos(5πx)
15π
1
− sin(27πt) cos(9πx) .
3π
72 A. K. MAJEE

6.5.1. Non-homogeneous wave equation in the interval [0, L]: Consider the non-homogeneous
wave equation with Dirichlet boundary conditions:

2
utt − c uxx = F (x, t), 0 < x < L, t > 0,

u(0, t) = f1 (t), u(L, t) = f2 (t), t > 0,

u(x, 0) = f (x), ut (x, 0) = g(x), 0 ≤ x ≤ L .


Suppose u is a solution of the above problem. Define

 x x
z(x, t) := u(x, t) − 1 − f1 (t) − f2 (t).
L L
Then, z satisfies the the non-homogeneous wave equation with homogeneous Dirichlet boundary
conditions.

2
ztt − c zxx = G(x, t), 0 < x < L, t > 0,

z(0, t) = z(L, t) = 0, t > 0,

z(x, 0) = v(x), zt (x, 0) = w(x), 0 ≤ x ≤ L ,


for some functions v(x), w(x) and G(x, t). It is clear that, if we solved for z, then we will get
the solution u of the given PDE.
How to find the solution z: Suppose we look for a solution of the type (as homogeneous
Dirichlet boundary conditions)
X nπ
z(x, t) = zn (t) sin( x).
L
n≥1

Then, we have
X c2 n2 π 2

nπ
ztt − c2 zxx = zn00 (t) + z n (t) sin( x).
L2 L
n≥1
Let us write X nπ
G(x, t) = Gn (t) sin( x).
L
n≥1
Then, we must have
c2 n2 π 2
Gn (t) = zn00 (t) + zn (t) . (6.12)
L2
Also, we need
z(x, 0) = v(x), and zt (x, 0) = w(x).
X nπ X nπ
If v(x) = vn sin( x) and w(x) = wn sin( x), then we should have
L L
n≥1 n≥1

zn (0) = vn , zn0 (0) = wn . (6.13)

Solving (6.12) and (6.13), we have unique solution zn (t). Hence
X nπ  x x
u(x, t) = zn (t) sin( x) + 1 − f1 (t) + f2 (t)
L L L
n≥1

is the unique solution of the given non-homogeneous wave equation with Dirichlet boundary
conditions.
Example 6.11. Let us consider the following PDE:

utt − uxx = et , 0 < x < 1, t > 0

u(0, t) = 0 = u(1, t), t>0
 PDES 73

u(x, 0) = x(x − 1), 0≤x≤1

ut (x, 0) = 0, 0≤x≤1
From the boundary conditions u(0, t) = u(1, t) = 0, it is clear that we should look for a solution
in terms of a Fourier sine series. The Fourier sine series of F (x, t) = et is given by (for n ≥ 1):
Z 1
2(1 − (−1)n ) t
Fn (t) = 2 F (x, t) sin(nπx) dx = e .
0 nπ
Thus, the Fourier series for et is given by
∞
X 2(1 − (−1)n )
et = sin(nπx) .
nπ
n=1

The Fourier sine series for f (x) = x(x − 1) is given by

∞
X 4((−1)n − 1)
x(x − 1) = sin(nπx).
(nπ)3
n=1
Substitute
∞
X
u(x, t) = un (t) sin(nπx)
n=1
into the equation utt − uxx = et , we obtain
∞ ∞
X X 2(1 − (−1)n ) t
u00n (t) + n2 π 2 un (t) sin(nπx) =


e sin(nπx) .
nπ
n=1 n=1
Thus, for n ≥ 1 and even, we get
u00n (t) + n2 π 2 un (t) = 0 =⇒ un (t) = Cn cos(nπt) + Dn sin(nπt).
Since n is even, the nth Fourier coefficient of f (x) is 0. Thus, we get that Cn = 0. Further,
since g(x) = 0, the nth Fourier coefficient is 0. Thus, we get that Dn = 0. We conclude that
un (t) = 0 for n ≥ 1 and even.
For n ≥ 1 and odd, we get
4 t
u00n (t) + n2 π 2 un (t) =
e.
nπ
Assume a particular solution of the form un (t) = cet . Then,
u00n (t) = cet and n2 π 2 un (t) = n2 π 2 cet .
Thus, we have,
u00n (t) + n2 π 2 un (t) = cet + n2 π 2 cet = c(1 + n2 π 2 )et .
Setting this equal to the right-hand side of the original equation:
4 t 4 4
c(1 + n2 π 2 )et = e =⇒ c = = .
nπ nπ(1 + n2 π 2 ) n(n2 π 2 + 1)π
Therefore, a particular solution is:
4
un,p (t) = et .
n(n2 π 2 + 1)π
The general solution is given by:
4
un (t) = Cn cos(nπt) + Dn sin(nπt) + et .
n(n2 π 2 + 1)π
74 A. K. MAJEE

Let us now use the initial condition to determine the constants Cn and Dn . In the case n ≥ 1
8
odd, we have the Fourier coefficient of x(x − 1) is − (nπ)3 . Thus, we get

4 8
Cn + =− .
n(n2 π 2
+ 1)π (nπ)3
The nth Fourier coefficient of g is 0, and so we get
4
u0n (0) = nπDn + 2 2
= 0.
n(n π + 1)π
Thus, the solution we are looking for is given by
X
u(x, t) = u2n+1 (t) sin((2n + 1)πx),
n≥0

where un (t), Cn , and Dn are given as above.

Remark 6.5. For non-homogeneous wave equation with Neumann boundary conditions

2
utt − c uxx = F (x, t), 0 < x < L, t > 0,

ux (0, t) = f1 (t), ux (L, t) = f2 (t), t > 0,

u(x, 0) = f (x), ut (x, 0) = g(x), 0 ≤ x ≤ L ,


one needs to define the function

x2 x2
 
z(x, t) := u(x, t) − x − f1 (t) − f2 (t)
2L 2L
so that z satisfies the non-homogeneous wave equation with homogeneous Neumann boundary
conditions

2
ztt − c zxx = G(x, t), 0 < x < L, t > 0,

zx (0, t) = zx (L, t) = 0, t > 0,

z(x, 0) = v(x), zt (x, 0) = w(x), 0 ≤ x ≤ L ,


for some functions v(x), w(x) and G(x, t). Therefore, we look for solution z of the form
X nπ
z(x, t) = Zn (t) cos( x)
L
n≥0

and determine Zn (t) uniquely from the given initial-boundary conditions.

Domain of dependence: We define the domain of dependence of the point (x0 , t0 ) as the
set
(
{x : |x − x0 | ≤ t0 }, n = 1 or n-even,
D(x0 , t0 ) :=
{x : |x − xo | = t0 } n ≥ 3 and n-odd.
Range of influence: The range of influence of a point x0 ∈ Rn is defined as the forward
light cone
(
{(x, t) : |x − x0 | ≤ t}, n = 1 or n-even,
R(x0 ) :=
{(x, t) : |x − x0 | = t}, n ≥ 3 and n-odd.

References
[1] Lawrence C. Evans. Partial Differential Equations.
[2] Robert C. McOwen. Partial Differential Equations: Methods and Applications.
[3] Fritz John. Partial Differential Equations.