Chapter 2

Partial differential equations

Symmetrical systems and partial differential equations of order I

Let D be an open set in Rn , and vi : D −→ R a collection of function in

the class C 0 (D), with 1 ≤ i ≤ n. We consider first a symmetrical system of
differential equations.

Definition 2.1 A symmetrical system of differential equations has the form

 dx
 dt1 = v1 (x1 , ..., xn )
··· , (2.1)
 dxn
dt = v (x
n 1 , ..., x n )

with t in an open interval I ∈ R .

At each point (x1, ..., xn ) ∈ D we can consider the vector

v̄ = (v1 (x1 , ..., xn ), ..., vn (x1 , ..., xn )),

so we get a vector field denoted also by v̄, in the class C 0 (D).

From (2.1) we also get that
dx1 dxn
= ... = = dt. (2.2)
v1 vn

Definition 2.2 A prime integral of the system (2.1) is a function ϕ : D −→ R

such that
∂ϕ dx1 ∂ϕ dxn
Dv̄ ϕ = · + ... + ·
dx1 dt dxn dt
dϕ
= (x1 (t), ..., xn (t)) = 0.
dt

33
34

As a direct consequence, we get

ϕ(x1 (t), ..., xn (t)) = c, (2.3)

with c ∈ R is a constant.

Definition 2.3 The curves t 7−→ (x1 (t), ..., xn (t)), defined by the condition
(2.1) or (2.3) are called integral curves.

We remark that the prime integrals are constant along the integral curves.

Definition 2.4 The prime integrals (ϕ1 , ..., ϕn ) are called functionally inde-
pendent on D if  
∂ϕ1 ∂ϕ1
· · ·
 ∂x1 ∂xn
rank  · · · · · · · · ·  = n.

∂ϕn
∂x1 · · · ∂ϕ
∂xn
n

Theorem 2.1 Let v̄ be a vector field on D which verifies the system (2.1).
Then, for any fixed t0 ∈ I, in a sufficiently small neighborhood of (x1 (t0 ), ...,
xn (t0 )) ∈ D there exist n − 1 prime integrals ϕ1 ,ϕ2 ,...,ϕn−1 , and any other
prime integral f is functionally dependent on {ϕi }i , 1 ≤ i ≤ n − 1, otherwise
said there is Φ : Rn−1 −→ R, with Φ ∈ C 1 (Rn−1 ) such that

ϕ = Φ(ϕ1 , ..., ϕn−1 ).

A convenient way for finding prime integrals is the method of integral

combinations.
We search for functions m1 , ..., mn which fulfil the following two conditions
1) ni=1 mi vi = 0,
P

2) there is a function ϕ such that m1 dx1 + ... + mn dxn = dϕ.

Under the above conditions, ϕ is a prime integral. Indeed, let us assume

that {xi (t)}i is a solution of (2.1). Then,

dϕ(x1 , ..., xn ) = m1 dx1 + ... + mn dxn

dx1 dxn
= m1 dt + ... + mn dt
dt dt
= m1 v1 dt + ... + mn vn dt
= (m1 v1 + ... + mn vn ) dt = 0,

so ϕ(x1 , ..., xn ) = c, with c ∈ R a constant, thus ϕ is a prime integral.

 35

Remark 2.1 Solving the system (2.1) amounts to finding exactly n−1 linearly
independent prime integrals. The general solution then will be of the form
ϕ = Φ(ϕ1 , ..., ϕn−1 ), Φ ∈ C 1 (Rn−1 ).

Now we introduce the definition of homogeneous partial differential equa-

tions of order I.

Definition 2.5 If {vi }i are functions in the class C 1 (D), then a homogeneous
partial differential equation of order I has the form
∂f ∂f
v1 + ... + vn =0 (2.4)
dx1 dxn

where the solution f : D −→ R should be clearly in the class C 1 (D).

Remark 2.2 There is an interplay between (2.2) and (2.4). To a partial

differential equation of order I we can attach a canonical symmetrical system
dx1 dxn
= ... = ,
v1 vn
such that the corresponding prime integral will be the solutions of (2.4).

Then we can solve homogeneous partial differential equation of order I

using the theory of symmetrical systems.
In the following we consider the case of non-homogeneous partial differen-
tial equation of order I.

Definition 2.6 A non-homogeneous partial differential equation of order I

has the form
∂f ∂f
v1 (x1 , ..., xn , f ) + ... + vn (x1 , ..., xn , f ) = g(x1 , ..., xn , f ) (2.5)
dx1 dxn

where vi : U −→ R, with U ⊆ Rn+1 for 1 ≤ i ≤ n, vi ∈ C 1 (U ), such that {vi }i

do not vanish simultaneously at any point.

In order to solve (2.5), we search the solution in the implicit form

F (x1 , ..., xn , f ) = 0,
∂F
with F ∈ C 1 (U ) and ∂f 6= 0. We obtain

∂F
∂f ∂x
= − ∂Fi ,
∂xi ∂f
36

for 1 ≤ i ≤ n. Plugging this into (2.5) we get another equation with solution
F , this time homogeneous:

∂F ∂F ∂F
v1 (x1 , ..., xn , f ) + ... + vn (x1 , ..., xn , f ) + g(x1 , ..., xn , f ) = 0,
dx1 dxn df

which can be solved as above, using the system

dx1 dxn df
= ... = =
v1 v1 g

If ϕ1 ,...,ϕn are functionally independent prime integrals of the system of

differential equations, then again the general solution will be of the form

ϕ = Φ(ϕ1 , ..., ϕn ),

with Φ ∈ C 1 (Rn ).

Partial differential equations of order II

In the sequel we will consider the following type of differential equations.

Let us consider Ω ∈ R2 an open and connected domain.

Definition 2.7 A quasi-linear partial differential equation of order II has the

following form

∂2u ∂2u ∂2u ∂u ∂u

A(x, y) 2
+ 2B(x, y) + C(x, y) 2
+ D(x, y, , , u) = 0. (2.6)
∂x ∂x∂y ∂y ∂x ∂y

The real functions A, B, C are assumed to be continuous; we also assume

that the functions do not vanish all at a point. The real function D is also
considered to be continuous in its variables. A function u = u(x, y, t) which
satisfies the equation (2.6) is called a solution.
We consider the following classification of the above defined differential
equations. If B 2 − AC > 0 on Ω, then it is called hyperbolic. If B 2 − AC = 0
it is parabolic, and if B 2 − AC < 0 then it is elliptic.
We sought to simplify the equation above by using a change of variables

ξ = ξ(x, y)
,
η = η(x, y)

such that the equation will be written as

∂2u ∂2u ∂2u ∂u ∂u

Ã(ξ, η) + 2 B̃(ξ, η) + C̃(ξ, η) + D̃(ξ, η, , , u) = 0. (2.7)
∂ξ 2 ∂ξ∂η ∂η 2 ∂ξ ∂η
 37

In general, studying the way the above function coefficients change when
changing the variables, it turns out that functionally independent prime inte-
grals of the ordinary differential equation
A(y 0 )2 − 2By 0 + C = 0, (2.8)
are the best solution, where y = y(x). The new partial differential equation
(2.7) is called the characteristic form of the initial equation, while the or-
dinary equation (2.8) is called the characteristic equation associated (or the
characteristic).
If the equation is hyperbolic, then the characteristic equation has two dis-
tinct real solutions, and we are able to get two prime integrals which will
determine the new variables ξ and η. In this case we will get vanishing coeffi-
cients à and C̃.
If the equation is parabolic, then the characteristic equation has a single
real solution which gives us the new variable ξ. For the second variable η we
may choose a simple expression functionally independent with respect to ξ,
usually η = x or η = y. In this case the new coefficients à and B̃ will vanish.
Finally, if the equation is elliptic, then the characteristic equation has two
conjugated complex roots. The real and, respectively the complex part will
give us the new variables ξ and η, respectively. The equation will have the
2 2
canonical form ∂∂ξu2 + ∂∂ηu2 + D̃(ξ, η, ∂u ∂u
∂ξ , ∂η , u) = 0.
For a partial differential equation of this type, of particular importance are
the initial conditions and boundary conditions. The first are conditions which
refer to the behavior of u at t = 0, while the later refer to the behavior of u
at the boundary of Ω.
We list now several classical problems which can be modeled using partial
differential equations.
Consider first the vibration of an infinite string, knowing the initial position
of the string and the initial velocity of each point. If the C 2 function u(t, x)
describes the position of the point whose projection on the horizontal axis is x
at the time t, then we can attach the partial differential equation with initial
conditions  ∂2u 2
 ∂t2 = a2 ∂∂xu2 ,
u(x, 0) = f (x),
 ∂u
∂t (x, 0) = g(x).
where x ∈ [0, l], t ≥ 0. Here the constant a is related to the elasticity of the
string, f (x) ∈ C 2 is the initial position on the vertical axis, g(x) ∈ C 1 being
the velocity.
Then the solution of the above differential equation is due to D’Alembert,
and can be written as
Z x+at
f (x − at) + f (x + at) 1
u(x, t) = + g(s)ds.
2 2a x−at
38

If the string has a finite length l, and fixed ends, then the attached equation
has initial condition and also boundary conditions (being also called mixed
problem)
 ∂2u 2

 ∂t2
= a2 ∂∂xu2 ,
 u(0, t) = 0,


u(l, t) = 0,
u(x, 0) = f (x),




 ∂u
∂t (x, 0) = g(x).

The solution of this problem can be obtained using the separation of vari-
ables, and can be written using Fourier series
∞ 
X nπa nπa  nπ
u(x, t) = An cos t + Bn sin t sin x,
l l l
n=1

where
Z l
2 nπ
An = f (x) sin xdx,
l 0 l
Z l
2 nπ
Bn = g(x) sin xdx.
nπa 0 l

Next, we consider a homogeneous rod of finite length, and let u(x, t) denote
now the temperature of the point with projection on the horizontal axis x at
the time t. Assume the temperature at the limiting points is 0. Then the
corresponding equation is

∂2u 2
a2 ∂∂xu2 ,


 ∂t2
=
u(0, t) = 0,


 u(l, t) = 0,

u(x, 0) = f (x),

with f ∈ C 2 .
Then, using again the Fourier method and the separation of variable, the
above heat problem has the solution
∞
n2 π 2 a 2 nπ
An e − t
X
u(x, t) = l2 sin x,
l
n=1

where Z l
2 nπ
An = f (x) sin xdx.
l 0 l
 39

Finally, consider the Dirichlet problem on a disk D := {(x, y) ∈ R2 |

x2 + y 2 ≤ r2 }, namely
( 2 2
∂ u
∂2x
+ ∂∂ 2uy = 0, for (x, y) ∈ D
u(x, y) = f (x) for (x, y) ∈ ∂D

where ∂D = {(x, y) ∈ R2 | x2 + y 2 = r2 }. To obtain the solution we pass to

polar coordinates 
x = ρ cos θ,
y = ρ sin θ,
with ρ ∈ (0, r] and θ = [0, 2π). Then
∞
X
u(ρ, θ) = A0 + (An cos nθ + Bn sin nθ) ρn ,
n=1

where the Fourier coefficients are

Z 2π
1
A0 = f (s)ds,
2π 0
Z 2π
1
An = f (s) cos nsds,
πrn 0
Z 2π
1
Bn = f (s) sin nsds.
πrn 0

2.1 Problems and exercises with solutions

1. Find the integral curves of the vector field associated to the symmetrical
system  dx1
dt = x2 ,
dx2
dt = −x1 .

Solution: We remark that the associated vector field is (x2 , −x1 ). We

differentiate the first equation, then use the second, and we obtain the
differential equation of order II
d2 x1
+ x1 = 0.
dt2
The characteristic polynomial is

P (X) = X 2 + 1,

with the roots λ1,2 = ±i, so the solution will be

x1 = c1 cos t + c2 sin t.
40

Differentiating again, we get the expression for x2 , so the integral curves

are 
x1 = c1 cos t + c2 sin t,
x2 = −c1 sin t + c2 cos t.

A prime integral can be also obtained, as we get

x21 + x22 = (c21 + c22 )(cos2 t + sin2 t)

= c21 + c22 = c,

where c is a real constant, so

ϕ(x1 , x2 ) = x21 + x22

is a prime integral. It it easy to see that the integral curves are circles
centered at the origin (see Figure 1 below).

R2 x2
v̄ = (x2 , −x1 )

x1

Figure 1

2. Solve the symmetrical system

dx dy dz
= 2 = .
x−z z 2z
Solution: We search for prime integrals. Firstly we consider the differ-
ential equation
dy dz
2
= .
z 2z
It gets
dy dz
= ,
z 2
 41

and we obtain
2dy = zdz.
We now integrate
z2
2y = + c,
2
so we can consider a first prime integral

z2
ϕ1 (x, y, z) = y − .
4
For the second prime integral we start with the remaining part of the
equation
dx dz
= ,
x−z 2z
which is equivalent to
dx x−z 1x 1
= = − .
dz 2z 2z 2
x
Let us denote z = u, equivalent to x = zu, and after differentiation it
becomes
dx du
=u+z .
dz dz
Using this, we have
du 1 1
u+z = u− ,
dz 2 2
and
du 1
z = − (u + 1).
dz 2
In the above differential equation we separate the variable
du dz
−2 = .
u+1 z
We now integrate and obtain

−2 ln(u + 1) = ln z + c.

Taking exponentials, this means

z(u + 1)2 = c.

Replacing u by xz , the second prime integral are

(x + z)2
ϕ2 (x, y, z) = .
z
42

We check now the functionally independence of the two prime integrals.

The Jacobi matrix is
!
∂ϕ1 ∂ϕ1 ∂ϕ1
∂x ∂y ∂z
J = ∂ϕ1 ∂ϕ1 ∂ϕ1
∂x ∂y ∂z
− z2
 
0 1
= 2(x+z) z 2 −x2 ,
z 0 z2

and the rank is clearly 2 (maximal), and the prime integrals are inde-
pendent.

3. Solve the partial differential equation of order I

∂f ∂f ∂f
(z + y) + (x + z) + (x + y) = 0.
∂x ∂y ∂y

Solution: The associated symmetrical system is

dx dy dz
= = .
z+y x+z x+y

For the first prime integral we can consider the relation

dx − dy dy − dz
= ,
y−x z−y

which can be derived form the equation. We write it in the following

convenient form
d(x − y) d(y − z)
= ,
−(x − y) −(y − z)
and after integration we get ln(x−y) = ln(y −z)+c; taking exponentials,
this gives
x−y
= c,
y−z
x−y
The prime integral can be considered as ϕ1 (x, y, z) = y−z .
For the second prime integral we can derive the following equation

dx − dy dx + dy + dz
= .
y−x 2(x + y + z)

The equivalent form is

d(x − y) d(x + y + z)
= .
−(x − y) 2(x + y + z)
 43

We integrate and obtain

1
− ln(x − y) = ln(x + y + z) + c,
2
equivalent with
1
= (x + y + z) · c,
(x − y)2
so the second prime integral can be considered

ϕ2 (x, y, z) = (x − y)2 (x + y + z).

It is an easy exercise to prove the functionally independence.

4. Find the general solution for the nonhomogeneous equation

∂z ∂z
2y + 3x2 = −6x2 y.
∂x ∂y

Solution: We associate the symmetrical system

dx dy dz
= 2 = .
2y 3x −6x2 y

From the first equality we obtain

3x2 dx = 2ydy,

so we can integrate and obtain

x3 − y 2 = c,

Then the first prime integral is ϕ1 (x, y, z) = x3 − y 2 .

For the second, consider
dz
dx = ,
−3x2
1
which is obtained if we factor out 2y , so −3x2 dx = dz. Then by integra-
tion, it gives x3 + z = c, the second prime integral being ϕ2 (x, y, z) =
x3 + z, which is clearly functionally independent from the first. Conse-
quently, the general solution will be of the form

ϕ = Φ(x3 − y 2 , x3 + z).
44

5. Solve the following homogeneous first order partial differential equations

∂f ∂f ∂f
2(y + z) +y +z = 0,
∂x ∂y ∂z
with the condition f (0, y, z) = y 2 + z 2 .
Solution: The canonically associated symmetrical system is
dx dy dz
= = .
2(y + z) y z
dy dz
From the last equality we obtain y = z , so
ln y = ln z + c̄.
y
Taking exponentials, this is z = c, c = ec̄ , so ϕ1 (x, y, z) = yz .
For the second prime integral, we have
dx d(y + z)
= .
2(y + z) y+z
From here, x − 2(y + z) = c, which gives us the second prime integral,
ϕ2 (x, y, z) = x − 2(y + z). The Jacobi matrix is
!
∂ϕ1 ∂ϕ1 ∂ϕ1
∂x ∂y ∂z
J = ∂ϕ1 ∂ϕ1 ∂ϕ1
∂x ∂y ∂z
0 z1 − zy2
 
= ,
1 −2 −2
which clearly has maximal rank. The general solution is
y
f (x, y, z) = Φ( , x − 2(y + z)).
z
The additional condition implies
y
Φ( , −2(y + z)) = y 2 + z 2 .
z
Denoting
ξ = yz

,
η = −2(y + z)
we get that (
ξη
y = − 2(ξ+1)
η ,
z = − 2(ξ+1)
2 2
and Φ(ξ, η) = η4(ξ+1)
(ξ +1) y
2 . Then, replacing now ξ by z and η by x−2(y+z),

the solution is
(x − 2(y + z))2 (y 2 + z 2 )
f (x, y, z) = .
4(y + z)
 45

6. Solve the partial differential equation

p
∂u ∂u x2 + y 2 ∂u
x +y + =0
∂x ∂x z ∂x

Solution: We associate the canonical symmetrical system

dx dy zdz
= =p .
x y x2 + y 2

From here we have the following more convenient form

xdx ydy zdz

2
= 2 =p .
x y x2 + y 2

By addition, it turns out that

xdx + ydy zdz

2 2
=p .
x +y x2 + y 2

Then
xdx + ydy
p = zdz.
x2 + y 2
This is in fact p
d( x2 + y 2 ) = zdz.
The later equation can be integrated, so we obtain
p z2
x2 + y 2 − = c,
2
p 2
so the first prime integral is ϕ1 (x, y, z) = x2 + y 2 − z2 . In order to
obtain the second prime integral, we consider the first equality

dx dy
= .
x y

Taking integrals, this gives

ln x = ln y + c̄,

and x = y · c, with c = ec̄ , after exponentiating the relation. So the

second prime integral is ϕ2 (x, y, z) = xy , which is clearly functionally
independent.
46

7. Solve the symmetrical system

dx dy dz
= = .
x2 + y 2 2xy xz
Then find the surface tangent to the associated vector field which con-
tains the curve  2
x − y2 = 1
Γ: .
z = 1
Solution: From the first equality we get
dx − dy dy + dy
= 2 ,
x2 + y 2 − 2xy x + y 2 + 2xy
which is
d(x − y) dy + dy
= .
(x − y)2 (x + y)2
By integration, this is
1 1
− = c1 ,
x+y x−y
2x
and the first integral is ϕ1 (x, y, z) = x2 −y 2
.
From the second equality
dy dz
= ,
2y z
which by integration is
1
ln y = ln z + c̄2 .
2
Taking exponentials, it gives
y
= c2 ,
z2
y
with c2 = ec̄2 . Then ϕ2 (x, y, z) = z2
.
For the second requirement, we actually have the conditions
 2x
2 2 = c1
 xy −y


z2
= c2

 z = 1
 2
x − y2 = 1

From here 2y = c1 , y = c2 , so

c1 − 2c2 = 0.
 47

But this means

2x y
− 2 2 = 0,
x2 − y 2 z
and the above condition can be rewritten as

x2 = y 2 + z 2 .

The above equation describes a cone.

8. Find the canonical form for the following equation of order II.

∂2u ∂2u ∂2u

+ 2 − 3 = 0.
∂x2 ∂x∂y ∂y 2
Solution: We consider the characteristic equation

(y 0 )2 − 2y 0 − 3 = 0.

We solve this classical equation of order 2 computing first the discrimi-

nant ∆ = 4 + 12 = 16. Let us remark that our equation is of hyperbolic
type. Then the solutions are

0 2±4
y1,2 = .
2

For y 0 = 3, we get
dy
= 3,
dx
so dy = 3dx, with the solution y = 3x + c1 , c1 ∈ R a constant. Then
3x − y = c1 , and a prime integral shows up

ϕ1 (x, y) = 3x − y.

For y 0 = −1, the ordinary differential equation is

dy
= −1,
dx
which means dy = −dx, with the solution y = −x+c2 , c2 ∈ R a constant.
Then x + y = c2 , and we also obtain a prime integral

ϕ2 (x, y) = x + y.

Now we can perform a change of variables

ξ = 3x − y,
η = x + y.
48

The partial derivatives are

(
∂ξ ∂ξ
∂x =3 ∂y = −1
∂η ∂η .
∂x =1 ∂y =1

We express the partial derivatives of the functions u with respect to the

initial variables using the new coordinates (ξ, η).
∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂x ∂ξ ∂x ∂η ∂x
∂u ∂u
= 3 + .
∂ξ ∂η

∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂y ∂ξ ∂y ∂η ∂y
∂u ∂u
= − + .
∂ξ ∂η

We pass now to partial derivatives of order II

∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= 3 · + 3 ·
∂x2 ∂ξ 2 ∂x ∂ξ∂η ∂x
∂ 2 u ∂ξ ∂ 2 u ∂η
+ · + 2·
∂ξ∂η ∂x ∂η ∂x
∂2u ∂2u ∂2u
= 9 2 +6 + 2.
∂ξ ∂ξ∂η ∂η

For the mixed term we obtain

∂2u ∂ 2 u ∂ξ ∂2u ∂η
= − 2
· − ·
∂x∂y ∂ξ ∂x ∂ξ∂η ∂x
∂ 2 u ∂ξ ∂2u ∂η
+ · + 2 ·
∂ξ∂η ∂x ∂η ∂x
∂2u ∂2u 2
∂ u
= −3 2 + 2 + .
∂ξ ∂ξ∂η ∂η 2

Finally,
∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= − · − ·
∂y 2 ∂ξ 2 ∂y ∂ξ∂η ∂y
∂ 2 u ∂ξ ∂ 2 u ∂η
+ · + 2·
∂ξ∂η ∂y ∂η ∂y
2
∂ u ∂2u ∂2u
= − 2 + .
∂ξ 2 ∂ξ∂η ∂η 2
 49

Plugging the above relations back in the initial equation allow us to

rewrite it as
∂2u ∂2u ∂2u
9 + 6 +
∂ξ 2 ∂ξ∂η ∂η 2
∂2u ∂2u ∂2u
 
+ 2 −3 2 + 2 +
∂ξ ∂ξ∂η ∂η 2
 2
∂2u ∂2u

∂ u
+2 − 2 + = 0,
∂ξ 2 ∂ξ∂η ∂η 2

or, after computations

∂2u
= 0,
∂ξ∂η
which represent the canonical form.

9. Find the canonical form for the following equation of order II.

∂2u ∂2u ∂2u

− 4 + 4 = 0.
∂x2 ∂x∂y ∂y 2

Solution: As before, we start out by considering the characteristic equa-

tion (the characteristic) canonically attached to the given PDE equation.

(y 0 )2 + 4y 0 + 4 = 0.

We computing the discriminant ∆ = 16 − 16 = 0. Consequently, in

contradistinction to the above example, our equation is parabolic. The
unique solution of the characteristic equation is

y 0 = −2.

We get
dy
= −2,
dx
so y = −2x + c, c ∈ R a constant. Then 2x + y = c, and a prime integral
is
ϕ1 (x, y) = 2x + y.

For the new variable η we can consider a function which is independent

to ϕ(x, y). We take η = x. The new variables are now

ξ = 2x + y,
η = x.
50

Compute first the partial derivatives

(
∂ξ ∂ξ
∂x = 2 ∂y =1
∂η ∂η .
∂x = 1 ∂y =0

Using the above relations we obtain the partial derivatives of the func-
tions u with respect to the initial variables using the new coordinates
(ξ, η).

∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂x ∂ξ ∂x ∂η ∂x
∂u ∂u
= 2 + .
∂ξ ∂η

∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂y ∂ξ ∂y ∂η ∂y
∂u
= .
∂ξ

Secondly, we compute the partial derivatives of order II

∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= 2 · + 2 ·
∂x2 ∂ξ 2 ∂x ∂ξ∂η ∂x
∂ 2 u ∂ξ ∂ 2 u ∂η
+ · + 2·
∂ξ∂η ∂x ∂η ∂x
∂2u ∂2u ∂2u
= 4 2 +4 + 2.
∂ξ ∂ξ∂η ∂η

For the mixed term we obtain

∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= − · + ·
∂x∂y ∂ξ 2 ∂x ∂ξ∂η ∂x
∂2u ∂2u
= 2 2 + .
∂ξ ∂ξ∂η

Finally,

∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= · + ·
∂y 2 ∂ξ 2 ∂y ∂ξ∂η ∂y
∂2u
= .
∂ξ 2
 51

Plugging the above relations into the initial equation allow us to rewrite
it as
∂2u ∂2u ∂2u
4 + 4 +
∂ξ 2 ∂ξ∂η ∂η 2
 2
∂2u
  2 
∂ u ∂ u
−4 2 2 + +4 = 0,
∂ξ ∂ξ∂η ∂ξ 2
or, after computations
∂2u
= 0,
∂η 2
which is the canonical form.
10. Find the canonical form for the following equation of order II.

∂2u ∂2u ∂2u

+ 4 + 5 = 0.
∂x2 ∂x∂y ∂y 2
Solution: The characteristic is

(y 0 )2 − 4y 0 + 5 = 0.

We solve the equation of order 2; the discriminant ∆ = 16 − 20 = −4.

So, the equation is of elliptic type. Then the solutions are
0
y1,2 = 2 ± i.

For y 0 = 2 + i, we get
dy
= 2 + i,
dx
so dy = 2dx + idx, with the solution y = 2x + ix + c, with c this time
lying in C. Then
(2x − y) + ix = c.
For y 0 = 2 + i we obtain the conjugate equation

(2x − y) − ix = c.

Then we choose the new variables to be

ξ = 2x − y,
η = x.

The partial derivatives are

(
∂ξ ∂ξ
∂x =2 ∂y = −1
∂η ∂η .
∂x =1 ∂y =0
52

As before,
∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂x ∂ξ ∂x ∂η ∂x
∂u ∂u
= 2 + .
∂ξ ∂η

∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂y ∂ξ ∂y ∂η ∂y
∂u
= − .
∂η

We pass now to partial derivatives of order II

∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= 2 · +2 ·
∂x2 2
∂ξ ∂x ∂ξ∂η ∂x
∂ 2 u ∂ξ ∂ 2 u ∂η
+ · + 2·
∂ξ∂η ∂x ∂η ∂x
∂2u ∂2u ∂2u
= 4 2 +4 + 2.
∂ξ ∂ξ∂η ∂η

For the mixed term we obtain

∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= − · − ·
∂x∂y ∂ξ 2 ∂x ∂ξ∂η ∂x
∂2u ∂2u
= −2 2 − .
∂ξ ∂ξ∂η

Finally,

∂2u ∂ 2 u ∂ξ ∂ 2 u ∂η
= − · − ·
∂y 2 ∂ξ 2 ∂y ∂ξ∂η ∂y
∂2u
= .
∂ξ 2

Plugging the above relations back in the initial equation allow us to

rewrite it as
∂2u ∂2u ∂2u
4 + 4 +
∂ξ 2 ∂ξ∂η ∂η 2
∂2u ∂2u
   2 
∂ u
+4 −2 2 − +5 =0
∂ξ ∂ξ∂η ∂ξ 2
 53

or, after computations

∂2u ∂2u
+ 2 = 0,
∂ξ 2 ∂η
which represent the canonical form.

11. For the following partial differential equation of order II with variable
coefficients find the canonical form using a convenient change of vari-
ables.

∂2u 2
2∂ u
x2 − y = 0.
∂x2 ∂y 2

Solution: Let us consider as above the characteristic equation

x2 (y 0 )2 − y 2 = 0.

Then the solutions are

0 y
y1,2 =± .
x
Let us remark that our equation is of hyperbolic type.
For y 0 = xy , we get
dy y
= ,
dx x
so ln y = ln x + c̄1 , with the solution y = c1 x, c1 = ec̄1 ∈ R a constant.
Then xy = c1 , and a prime integral is
y
ϕ1 (x, y) = .
x

For y 0 = − xy , we obtain
dy y
=− ,
dx x
so ln y = − ln x + c̄2 ; the corresponding solution is y = c2 x1 , c2 = ec̄2 ∈ R
a constant. From here xy = c2 , and a prime integral is

ϕ2 (x, y) = xy.

Now we can perform a change of variables

y
ξ = ,
x
η = xy.
54

The partial derivatives are

(
∂ξ
∂x = − xy2 ∂ξ
∂y = x1
∂η ∂η .
∂x =y ∂y =x

We express the partial derivatives of the functions u with respect to the

initial variables using the new coordinates (ξ, η).
∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂x ∂ξ ∂x ∂η ∂x
y ∂u ∂u
= − 2 +y .
x ∂ξ ∂η

∂u ∂u ∂ξ ∂u ∂η
= · + ·
∂y ∂ξ ∂y ∂η ∂y
1 ∂u ∂u
= +x .
x ∂ξ ∂η

We compute now the partial derivatives of order II

∂2u y ∂ 2 u ∂ξ ∂ 2 u ∂η
 
y ∂u
= 2 3 − 2 · + ·
∂x2 x ∂ξ x ∂ξ 2 ∂x ∂ξ∂η ∂x
 2
∂ 2 u ∂η

∂ u ∂ξ
+x · + ·
∂ξ∂η ∂x ∂η 2 ∂x
y ∂u y 2 ∂ 2 u y2 ∂ 2u ∂2u
= 2 3 + 4 2 −2 2 + y2 2 .
x ∂ξ x ∂ξ x ∂ξ∂η ∂η

As in the initial equation the middle term is 0, we pass to the computa-

tion of the final, term

∂2u 1 ∂ 2 u ∂ξ ∂ 2 u ∂η ∂ 2 u ∂ξ ∂ 2 u ∂η
 
= · + · − · − ·
∂y 2 x ∂ξ 2 ∂y ∂ξ∂η ∂y ∂ξ 2 ∂y ∂ξ∂η ∂y
 2
∂ 2 u ∂η

∂ u ∂ξ
+x · + ·
∂ξ∂η ∂y ∂η 2 ∂y
1 ∂2u ∂2u 2∂ u
2
= + 2 + x .
x2 ∂ξ 2 ∂ξ∂η ∂η 2

Then we plug the above relations in initial equation and obtain after
computations
∂2u y ∂u
2y 2 − = 0,
∂ξ∂η x ∂ξ
 55

which is equivalent to
∂2u 1 ∂u
2 − = 0,
∂ξ∂η η ∂ξ
the later representing the canonical form.
12. Compute the solution of the following D’Alembert problem of a vibrating
string of infinite length.
 ∂2u 2
 ∂t2 − 9 ∂∂xu2 = 0,
u(x, 0) = x,
 ∂u
∂t (x, 0) = 3.

Solution: As explained in the introductory section of this chapter, the

solution of the above equation comes as a sum
u(x, t) = ϕ(x − 3t) + ψ(x + 3t).
From the boundary conditions we determine the functions ϕ and ψ.
More precisely, in our case
ϕ(x) + ψ(x) = x,
−3ϕ (x) + 3ψ 0 (x) = 3.
0

Then, we obtain
Z x+3t
1 1
u(x, t) = ((x − 3t) + (x + 3t)) + 3dz
2 6 x−3t
1
= x + · 6t · 3
6
= x + 3t.

13. Solve the following mixed problem of a vibrating string of finite length
π, with the corresponding boundary and initial conditions.
 ∂2u 2
− a ∂∂xu2 = 0,
 ∂t2


u(0, t) = u(π, t) = 0,

 u(x, 0) = sin x2 ,
 ∂u
∂t (x, 0) = cos x2 .
for 0 ≤ x ≤ π.
Solution: For the particular wave problem we have a solution which
comes as a Fourier series
∞
X
u(x, t) = (An cos nt + Bn sin nt) sin nx.
n=1
56

We compute the Fourier coefficients,

2 π
Z
x
An = sin sin nxdx
π 0 2
Z π    Z π    
1 1 1
= cos − n x dx − cos + n x dx
π 0 2 0 2
1 2  π  1 2  π 
= sin − nπ − sin + nπ
π 1 − 2n 2 π1 + 2n 2
2 1 1
= (−1)n −
π 1 − 2n 1 + 2n
(−1)n 8n
= .
π 1 − 4n2
Similarly,
Z π
2 x
Bn = cos sin nxdx
nπ 0 2
Z π    Z π    
1 1 1
= sin n+ x dx + sin n− x dx
nπ 0 2 0 2
1 −2   π 
= cos nπ + −1
nπ 2n + 1 2
1 −2   π 
+ cos nπ − −1
nπ 2n − 1 2
8 1
=− .
π 1 − 4n2

Then the solution will be

∞ 
8 X (−1)n n

1
u(x, t) = cos nt − sin nt sin nx.
π 1 − 4n2 1 − 4n2
n=1

14. Solve the following mixed problem for the heat equation of a rod of finite
length π, with the corresponding boundary and initial conditions.
 ∂u 2
 ∂t − a ∂∂xu2 = 0,
u(0, t) = u(π, t) = 0, for 0 ≤ x ≤ π.

u(x, 0) = sin x,

Solution: For this heat equation, we know that the solution has the
form
∞
2 2
X
u(x, t) = An e−n a t sin nx.
n=1
 57

We compute now the Fourier coefficients

2 π
Z
An = sin x sin nxdx
π 0
Z π Z π 
1
= cos ((1 − n) x) dx − cos ((1 + n) x) dx .
π 0 0

If n = 1, then clearly A1 = 1. If n > 1, then

1 sin (1 − n) x π sin (1 + n) x π
   
An =
π 1−n  − 1+n 
0 0
= 0.

Thus

2
u(x, t) = e−a t sin x.

15. Solve the following Dirichlet problem on the unit disk with the corre-
sponding boundary condition.
( 2 2
∂ u
∂x2
+ ∂∂yu = 0, for (x, y) ∈ D,
u(x, y) = y, for (x, y) ∈ ∂D.

where D = (x, y) ∈ R2 | x2 + y 2 ≤ 1 .


Solution: We pass to polar coordinates


x = ρ cos θ,
y = ρ sin θ.

with ρ ∈ (0, 1], θ ∈ [0, 2π).

Then, on the boundary y = 1 sin θ = sin θ.
As pointed out in the introductory section, the solution has the form
∞
X
u(x, t) = A0 + (An cos nθ + Bn sin nθ) ρn .
n=1

We compute the Fourier coefficients

Z 2π
1
A0 = sin θdθ
2π 0
2π
1 
= − cos θ = 0.
2π 0
58

Furthermore,

1 2π
Z
An = sin θ cos nθdθ
π 0
Z 2π Z 2π 
1
= sin ((n + 1) θ) dθ − sin ((n − 1) θ) dx .
2π 0 0

If n = 1, then clearly
Z 2π
1
A1 = sin 2θdθ
2π 0
2π
1 2 

= sin θ
2π 0
= 0.

If n > 1, then
 !
cos ((n + 1) θ) 2π cos ((n − 1) θ) 2π

1
An = −  +
2π n+1 0 n−1 
0

= 0.

For Bn , we get

1 2π
Z
Bn = sin θ sin nθdθ
π 0
Z 2π Z 2π 
1
= cos ((1 − n) θ) dθ − cos ((1 + n) θ) dx .
2π 0 0

If n = 1, then B1 = 1.
If n > 1, then

sin (1 − n) θ π sin (1 + n) θ π
   
1
Bn =
2π 1−n  − 1+n 
0 0
= 0.

Thus

u(ρ, θ) = sin θ · ρ.
 59

2.2 Problems and exercises

1. For the following hyperbolic partial differential equation compute the
canonical form.

∂2u ∂2u 2 ∂2u ∂u

2
+ 2 sin x − (3 + cos x) 2
−y = 0.
∂x ∂x∂y ∂y ∂y
Solution: We use the new coordinates ξ = cos x + 2x + y, η = cos x −
2x + y; the canonical form is

∂2u
 
ξ + η ∂u ∂u
+ 2 + = 0.
∂ξ∂η ξ ∂ξ ∂ξ

2. For the following parabolic partial differential equations compute the

canonical form.
2 2 2
∂ u
a) tg 2 x ∂∂xu2 −2ytg x ∂x∂y + y2 ∂∂yu2 + tg 3 x ∂u
∂x = 0

2 2 2
∂ u
b) y 2 ∂∂xu2 − 2xy ∂x∂y + x2 ∂∂yu2 − x ∂u ∂u
∂x − y ∂y = 0.
Solution: a) Performing the change of variables ξ = y sin x, η = y, we
obtain the canonical form
∂ 2 u 2ξ ∂u
− 2 = 0.
∂η 2 η ∂ξ

b) For ξ = x2 + y 2 , η = x, we obtain the canonical form

∂2u η ∂u
2
− = 0.
∂η ξ − η 2 ∂η

3. For the following elliptic partial differential equations compute the canon-
ical form.
∂2u 2
a) ∂x2
+y ∂∂yu2 + 1 ∂u
2 ∂x = 0, with y > 0.

∂2u 2
b) ∂x2
+ x ∂∂yu2 = 0, with x > 0.
∂2u 2 2
c) ∂x2
+8 ∂∂yu2 + 20 ∂∂yu2 + ∂u
∂x + 4 ∂u
∂y = 0,

√
Solution: a) Using the variables ξ = 2 y, η = x, we obtain the canon-
ical form
∂2u ∂2u
+ 2 = 0.
∂ξ 2 ∂η
60

√
b) For ξ = y, η = 32 x x, we obtain

∂2u ∂2u 1 ∂u
2
+ 2+ = 0.
∂ξ ∂η 3η ∂η

c) For ξ = y − 4x, η = 2x, we obtain

∂ 2 u ∂ 2 u 1 ∂u
+ 2+ = 0.
∂ξ 2 ∂η 2 ∂η

4. Integrate the following hyperbolic partial differential equations of order

II
2 2
a) 4x2 ∂∂xu2 −y 2 ∂∂yu2 + 2x ∂u
∂x = 0,

∂2u 2
∂ u 2
b) ∂x2
+ 2 cos x ∂x∂y − sin2 x ∂∂yu2 − sin x ∂u 4
∂y = 0, with u(x, sin x) = x ,
∂u
∂y (x, sin x) = x,
∂2u ∂2u
c) −
∂x2 ∂y 2
+ 3 ∂u ∂u x
∂x − 3 ∂y = e .

x
Solution: a) Performing the change of variables ξ = y2
, η = xy 2 , we
obtain the solution
 
p x
4
+ ψ xy 2 .


2
u(x, y) = xy ϕ
y 2

b) For ξ = x − y + sin x, η = −x − y + sin x, we obtain the solution

1h i
u(x, y) = (x + y − sin x)4 + (x − y + sin x)4
2
1h i
− (x + y − sin x)2 − (x − y + sin x)2
4

c) Performing the change of variables ξ = x + y, η = x − y, we obtain

the solution
ex
u(x, y) = ϕ (x + y) + e−3y ψ (x − y) + .
4

5. Find a particular solution for the partial differential equations of order

II.

∂ 2 u ∂u
− = 2 cos (3x − 2t) .
∂x2 ∂t
 61

Solution: We search a solution of the form

u(x, t) = a cos (3x − 2t) + b sin (3x − 2t) .

We find
4 18
u(x, y) = sin (3x − 2y) − cos (3x − 2y) .
85 85
6. For the following parabolic partial differential equations of order II find
the general solution.
∂2u ∂2u ∂u
a) ∂x2
+2 ∂x∂y + ∂x + 2 ∂u
∂y = 0.

2 2 ∂2u
∂ u
b) 4 ∂∂xu2 − 4 ∂x∂y + ∂y 2
− 6 ∂u ∂u
∂x + 3 ∂y − 4u = 2e
x−y .

Solution: a) Performing the change of variables ξ = y, η = y − 2x, we

obtain the solution
y
u(x, y) = e 2−x + ϕ (y) + ψ (y − 2x) .

b) Performing the change of variables ξ = x + 2y, η = y, the solution is

1
u(x, y) = ey ϕ (x + 2y) + e−4y ψ (x + 2y) − ex−y .
2

7. Determine the solution of the following partial differential equation.

 ∂2u 1 ∂2u
 ∂x2
− 4 ∂t2 = 0,
3 sin πx

 u(x, 0) = 4 ,


∂u
∂t (x, 0) = 0,



 u(0, t) = 0,

u(4, t) = 0.

Solution: u(x, t) = 3 sin πx πt

4 cos 2 .

8. Find the solution of the following partial differential equation.

 2 2

 a2 ∂∂xu2 − ∂∂t2u = 0,
 2h
0 ≤ x ≤ 2l

l x,


 u(x, 0) =


2h l
l (l − x) , 2 ≤ x ≤ l
∂u
∂t (x, 0) = 0






 u(0, t) = 0,
 u(l, t) = 0.

P∞ (−1)k
Solution: u(x, t) = 8h
π2 k=0 (2k+1)2 sin (2k+1)π
l x cos a(2k+1)π
l t.
62

9. Find the solution of the following partial differential equation, for l ≥ 0,

0 ≤ x ≤ l.  2 ∂ 2 u ∂u

 a ∂x2 − ∂t = 0,
 u(x, 0) = x2 − 1,


limt→∞ u(x, t) = 0,
 ∂u
∂t (0, t) = 0,



 ∂u
∂t (l, t) = 0.

(−1)k akπ 2
4l2 P∞ −( ) .
t
Solution: u(x, t) = π2 k=0 k2
cos kπ
l xe
l

10. Determine the solution of the following partial differential equation.

 2
∂ u ∂2u ∂2u
 ∂x2 + 2 ∂x∂y − 3 ∂y2 = 0,

u(x, 0) = 3x2 ,
 ∂u (x, 0)

= 0.
∂y

Solution: u(x, t) = 3x2 + y 2 .

11. Find the solution of the partial differential equation.


∂2u ∂2u ∂2u
 2 ∂x2 − 7 ∂x∂y + 5 ∂y2
 = 0,
u(0, y) = 9y 2 ,
 ∂u (0, y)

= y2.
∂y

Solution: We use the change of coordinates ξ = 3x + y, η = x + 2y.

The solution is
 
1 2 2 3 2 1 2
u(x, y) = (3x + y) − (3x + y) + (x + 2y) − (x + 2y)
5 4 4

12. Find the solution of the partial differential equation.

 2 2
 ∂∂t2u − ∂∂xu2 − ex = 0,

u(0, x) = sin x,
 ∂u (0, x)

= x + cos x.
∂y

1
Solution: u(x, t) = 3 sin x cos t + xt − ex + ex+t + ex−t .
 
2

13. Compute the solution of the partial differential equation, where 0 < x <
1. 
2 ∂2u ∂2u ∂u
 (1 − x ) ∂x2 − ∂y2 − x ∂x = 0,

u(x, 0x) = arcsin x,
 ∂u (x, 0)

= 1.
∂y
 63

Solution: We use the new variables ξ = arcsin x + y, η = arcsin x − y.

The solution is
u(x, y) = arcsin x + y.

14. Find the general solution of the following Cauchy problem

 2
∂ u ∂2u ∂2u
 ∂x2 − ∂x∂y − 6 ∂y2
 = 0,
u(0, y) = f (y),
 ∂u (0, x)

= g(y).
∂x

1 1
R y+2y
Solution: u(x, t) = 5 (3f (y − 2x) + 2f (y + 2x)) + 5 y−2x g(t)dt.

15. Find the general solution of the following Cauchy problem


∂2u ∂ u 2 2

 + 4 ∂x∂y
∂x2
+ 4 ∂∂yu2 = ex ,
u(0, y) = f (y),
 ∂u (0, x)

= g(y).
∂x

Solution: u(x, t) = f (y − 2x) + xg(y − 2x) + f 0 (y − 2x) + ex − x − 1.