Partial Differential Equations of First Order and Their Applications To Physics G Lopez PDF

January 22, 2014 14:45 World Scientific Book - 9in x 6in ws-book9x6˙x˙x
Publishers’ page
Publishers’ page
Publishers’ page
Publishers’ page
I want to dedicate this book to my parents, J. Guadalupe López López and

Vicenta Velázquez, for their support and their love during their lifetime.
I also want to dedicate this book to my sister Esperanza and my friend

Javier I. Hernández who both passed away and their life were an
inspiration to me.
I finally want to dedicate this book to my granddaughter Jezen for being

a wonderful and valuable gift to our life.
vi
Preface
This book is a collection of notes of the course Partial Differential Equa-

tions of First Order given by Dr. Gustavo López Velázquez during different
periods of time at different universities. The main body of this book was
put together during a seminar, given by Dr. G. López in collaboration with
M.A. Murguı́a and M. Romero at the Physics Institute of the University of
Guanajuato, from August to November of 1988. This collection was made
with the help of M.A. Murguı́a, M. Romero, E. Benitez, and C. Melo. The
final version of the first edition was made by Dr. López at the University of
Guadalajara, México, and at at Los Alamos National Laboratory (Depart-
ment of Non Linear Dynamics), New Mexico, USA, from 1995 to 1996. I
want to point out that without the help and enthusiasm of M.A. Murguı́a,
M. Romero and C. Melo, the elaboration of these notes would not have
been possible. In addition, I want to thank also to Alejandra Taylor for her
collaboration during the revision of the text.
For the second edition, I want to thank to Gustavo Montes for his help
for reviewing and typing much of the material of this new version. I ex-
panded the contain of each chapter adding new material which I considered
relevant for the students, I did a lot of typing corrections appearing on the
first edition, for which I apologize to the readers. Finally, I put several
exercises on each chapter which can help to master each topic.
In this book I try to point out the mathematical importance of Partial

Differential Equations of First Order in Physics and Applied Sciences. The
intention of this book is to give to mathematicians a wide view of the ap-
vii
viii Preface
plications of this branch in Physics , and to give to physicists and applied

scientists a powerful tool for solving some problems appearing in Classical
Mechanics, Quantum Physics,Optics, and General Relativity. This book is
intended for senior or first year graduated students in mathematics, physics
or engineering curricula.
Mathematics is a gift. . .
for man to understand the laws of nature
which make up the whole Universe.
G. López
Contents
Preface vii
1. Geometric Concepts and Generalities 1

1.1 Surfaces and Curves in Three Dimensions . . . . . . . . . 1
1.2 Parallelism of vector fields . . . . . . . . . . . . . . . . . . 9
1.3 Methods of Solution of dx/P = dy/Q = dz/R . . . . . . . 11
1.4 Orthogonal Trajectories of a System of Curves on a Surface 15
1.5 Pfaffian Differential Equation in R3 . . . . . . . . . . . . 17
1.6 N. Mechanics, Lagrangians, Hamiltonian, H-J Equation,
and Liouville’s Theorem . . . . . . . . . . . . . . . . . . . 19
1.6.1 Equivalent Hamiltonians . . . . . . . . . . . . . . 27
1.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2. Partial Differential Equations of First Order 33

2.1 Classification . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Linear PDEFO for Functions Defined in Ω ⊂ R2 . . . . . 35
2.3 Quasi-linear PDEFO for Functions Defined in Ω ⊂ R2 . . 39
2.4 Quasi-linear PDEFO for Functions Defined in Ω ⊂ Rn . . 45
2.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3. Physical Applications I 53
3.1 Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Angular Momentum in Quantum Mechanics . . . . . . . . 60
3.3 Heat Propagation between Two Superconducting Cables . 67
ix
x Contents
3.4 Classical Statistical Mechanics in Equilibrium . . . . . . . 71

3.5 Renormalization Group’s Equations . . . . . . . . . . . . 73
3.6 Particle Multiplicity Distribution in High Energy Physics 76
3.7 Hamiltonian Perturbation Approach in Accelerator Physics 79
3.8 Perturbation Approach for the One-dimensional Constant
of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.9 Constant of Motion for a Relativistic Particle under Peri-
odic Perturbation . . . . . . . . . . . . . . . . . . . . . . . 86
3.10 Uniqueness of constant of motion . . . . . . . . . . . . . 87
3.11 Vlasov Equation and Bunched Beam Instabilities . . . . 91
3.11.1 Potential well distortion . . . . . . . . . . . . . . . 92
3.11.2 Longitudinal Modes . . . . . . . . . . . . . . . . . 94
3.11.3 Transverse Modes . . . . . . . . . . . . . . . . . . 96
3.12 Interaction Plasma-Electromagnetic Field . . . . . . . . . 100
3.12.1 Diluted Plasma . . . . . . . . . . . . . . . . . . . 100
3.12.2 Diluted linear, dispersed and nonlocal plasma . . 105
3.12.3 Non Dilute Plasma . . . . . . . . . . . . . . . . . 107
3.13 Decoherence in a quantum system . . . . . . . . . . . . . 109
3.14 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.15 References . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4. Nonlinear Partial Differential Equations of First Order 127

4.1 Non-linear PDEFO for Functions Defined in R2 . . . . . . 127
4.2 Non-linear PDEFO for Functions Defined in Rn . . . . . . 143
4.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5. Physical Applications II 153

5.1 Motion of a Classical Particle . . . . . . . . . . . . . . . . 153
5.1.1 Hamilton-Jacobi equation for a one-dimensional
Harmonic Oscillator . . . . . . . . . . . . . . . . . 153
5.1.2 The Lagrangian obtained directly from Hamiltonian155
5.1.3 Relativistic particle moving in a Coulomb field . . 157
5.1.4 Motion of a test particle in a Schwarchild’s space 160
5.1.5 Interaction of a periodic gravitational wave with a
test particle . . . . . . . . . . . . . . . . . . . . . 163
5.2 Trajectory of a Ray of Light . . . . . . . . . . . . . . . . . 167
Contents xi
5.2.1 Solution of the Eikonal equation for a refraction

index depending on z . . . . . . . . . . . . . . . . 168
5.2.2 Solution of the Eikonal equation for a refraction
index radially depending . . . . . . . . . . . . . . 169
5.2.3 Eikonal equation for a refraction index radially de-
pending in sphere . . . . . . . . . . . . . . . . . . 170
5.3 Equivalent Hamiltonians . . . . . . . . . . . . . . . . . . . 170
5.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
6. Characteristic Surfaces of Linear Partial Differential

Equation of Second Order 177
n
6.1 Characteristic Surfaces of a Linear PDESO Defined in R 177
6.2 Characteristic Surfaces of a Linear PDESO Defined in R2 180
6.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Chapter 1
Geometric Concepts and Generalities
In this chapter we shall study some geometric concepts that are basic to
understand the geometric meaning of the partial differential equation in
R3 .
1.1 Surfaces and Curves in Three Dimensions
By a surface S in R3 we mean any relation between the rectangular carte-

sian coordinates (x, y, z) of a point in this space given by following expres-
sions
(explicit) z = f (x, y) (1.1)
(implicit) F (x, y, z) = 0 (1.2)
(parametric) x = f1 (u, v), y = f2 (u, v), z = f3 (u, v) (1.3)
where to each pair of values of u, v there corresponds a set of numbers
(x, y, z) and hence a point in space. While the expressions for the surfaces
(1.1) and (1.2) are unique, the parametric expression (1.3) is not unique.
For example, the spherical surface x2 + y 2 + z 2 = a2 , (see Fig. 1.1) can be
parameterized by
x = a sin u cos v, y = a sin u sin v, z = a cos u,
or
1 − v2 1 − v2 2ab
x=a cos u, y=a sin u, z= ,
1 + v2 1 + v2 1 + v2
or more general
p
1 − g(v) 1 − g(v) 2a g(v)
x=a cos u, y=a sin u, z= ,
1 + g(v) 1 + g(v) 1 + g(v)
1
2 Geometric Concepts and Generalities
Fig. 1.1 Surface of sphere is not unique parameterized.
where g(v) is such that 1 + g(v) > 0 for all v ∈ R. By a curve Γ in R3 we

understand any relation between a point (x, y, z) in this space of the form
(non-parametric) f (x, y, z) = 0; g(x, y, z) = 0 (1.4)
(parametric) x = f1 (t), y = f2 (t), z = f3 (t) (1.5)
where t is a continuous variable called the parameter of the curve (a usual
parameter is the length of a curve measured from some fixed point). The
relation (1.4) expresses in fact the intersection of two surfaces (see Fig.
1.2).
A surface can be thought of as being generated by a set of curves in the
following way, the surface f (x, y, z) = 0 is generated by set of curves Γk
defined by
x = k, f (x, y, k) = 0 (1.6)
where k takes a certain interval of values, for example, the sphere (see Fig.
1.3) can be seen as generated by the curves Γk given by
z = k, x2 + y 2 = a2 − k 2 with −a≤k ≤a .
Let a curve Γ be parameterized by the length of the curve s, an let
P0 = (x(0), y(0), z(0)),
P = (x(s), y(s), z(s)),

and
Q = (x(s + δs), y(s + δs), z(s + δs))
1.1. Surfaces and Curves in Three Dimensions 3
Fig. 1.2 The intersection of the surfaces S1 and S2 generated the curve Γ.
Fig. 1.3 Surface of sphere is generated by the planes z = k, −a ≤ k ≤ a.
be three points on Γ (see Fig. 1.4) . If δc is the Euclidean distance between

the point P and Q, we restrict ourselves to those kind of curves which
satisfy
δc
lim = 1. (1.7)
δs→0 δs
This means, for example that we will not be interested in such a curves
turn around and cross themselves in some point. The directions cosines of
Fig. 1.4 Distances δc and δs between the points P and Q.
the chord P Q are

x(s + δs) − x(s) y(s + δs) − y(s) z(s + δs) − z(s)
, , ,
δc δc δc
and due to the Taylor’s theorem,
x(s + δs) − x(s) = δs(dx/ds) + O(δs2 ),
these direction cosines are deduced to

δs dx dy dz
, , + O(δs2 ),
δc ds ds ds
and δs tends to zero. The chord P Q takes up the direction of the tangent
to the curve at P , and according to Eq. (1.7), the direction cosines of this
tangent are

dx dy dz
, , . (1.8)
ds ds ds
By a curve Γ given in parametric form , with s as the parameter, and
passing upon a surface S given by expression (1.2) (see Fig. 1.5) , we
understand that the following identity is satisfied
F (x(s), y(s), z(s)) = 0 (1.9)
for all the values s in the curve which lies on the surface.
If Eq. (1.9) is satisfied for all values of s, then the curve lies completely
on the surface. Of course, if the curve is caused by the intersection of two
surfaces, this curve lies completely on both surfaces. Differentiating Eq.
(1.9) with respect to s, we obtain
∂F dx ∂F dy ∂F dz
+ + = 0. (1.10)
∂x ds ∂y ds ∂z ds
Fig. 1.5 Part of a curve Γ laying on the sphere.
From the relation (1.8) and (1.10) we see that the tangent to the curve Γ
at any point P on the surface S is perpendicular to the gradient of F

∂F ∂F ∂F
∇F = , , , (1.11)
∂x ∂y ∂z
and this is true for any curve Γ lying on S passing through P , then the
vector ∇F is normal to the surface S at the point P (see Fig. 1.6) , and the
normal vector at any point of the surface is given by this gradient, valued
at this point, divided by its magnitude. If the equation of the surface S is
given in the form z = f (x, y), defining p and q as
∂z ∂z
p= , and q= , (1.12)
∂x ∂y
and making F = f (x, y) − z, it follows that Fx = p, Fy = q, Fz = −1 and
the unitary vector n̂ normal to the surface in any point is
1
n̂ = 2 (p, q, −1). (1.13)
[p + q 2 + 1]1/2
Let P = (x, y, z) be a point on the surface S defined by F (x, y, z) = 0 and
let π1 be the tangent plane at this point, if (X, Y, Z) is any other point on
π1 then, from the above discussion, the vector (X − x, Y − y, Z − z) lying
Fig. 1.6 Normal vector ∇F at the point P of the sphere, and the tangent T1 and
orthogonal T2 lines of to the curve Γ at this point.
on the plane π1 , must be perpendicular to the normal directions ∇F at P ,

so the equation tangent plane π1 , (see Fig. 1.7) is
∂F ∂F ∂F
(X − x) + (Y − y) + (Z − z) = 0. (1.14)
∂x ∂y ∂z
Similarly, let S2 be other surface defined by G(x, y, z) = 0 which intersects
the surface S1 generating a curve Γ that passes through the point P . The
equation for the tangent plane π2 of the surface at the point P is
∂G ∂G ∂G
(X 0 − x) + (Y 0 − y) + (Z 0 − z) = 0, (1.15)
∂x ∂y ∂z
where (X 0 , Y 0 , Z 0 ) is now any other point on this tangent plane π2 (see
Fig. 1.8) . The equation of the line L generated by the intersection of both
planes π1 and π2 must be such that its directions cosines vector (X 00 −
x, Y 00 − y, Z 00 − z), where (X 00 , Y 00 , Z 00 ) is now any other point on the line
L, is perpendicular to ∇F and ∇G, that is, it must be parallel to the cross
product of ∇F with ∇G,

∂F ∂G ∂F ∂G ∂F ∂G ∂F ∂G ∂F ∂G ∂F ∂G
∇F × ∇G = − , − , − ,
∂y ∂x ∂z ∂y ∂z ∂x ∂x ∂z ∂x ∂y ∂y ∂x
(1.16)
Fig. 1.7 Tangent plane π of the surfaces S and S2 at point P = (x, y, z), and vector
joining this point with other point, (X, Y, Z), on the plane.
and therefore is proportional to this vector, establishing the following equa-

tions
X 00 − x Y 00 − y Z 00 − z
= = , (1.17)
∂(F, G) ∂(F, G) ∂(F, G)
∂(y, z) ∂(z, x) ∂(x, y)
where ∂(F G)/∂(y, z) is given by

∂(F, G) Fy Fz
= det = Fy Gz − Fz Gy , (1.18)
∂(y, z) Gy Gz
and so on. Choosing the point on L close enough to (x, y, z), i.e. X 00 =
x + dx, Y 00 = y + dy, Z 00 = z + dz, and given F and G, then (1.17) has the
following form
dx dy dz
= = , (1.19)
P (x, y, z) Q(x, y, z) R(x, y, z)
where P , Q and R are known functions. The solution of
(1.19) gives us the lines with tangent parallel to the vector field
(P (x, y, z), Q(x, y, z), R(x, y, z)). These integral curves form a two-
parameter family of curves in three dimensional space.
√
Example√ 1.1. Give the tangent planes at the point P = (0, ( 17/2 −
1/2)1/2 , ( 17 − 1)/2) of the surface x2 + y 2 + z 2 = 4. Give their normal
vectors and the equation of the tangent line generated by the intersection
Fig. 1.8 Interception of the tangent planes π1 and π2 to the surfacesS1 and S2 at the
point P , generating the tangent line L.
of the planes at this point. Give the curve Γ generated by the intersection
of both surfaces.
In this case
F = x2 + y 2 + z 2 − 4 = 0
and
G = x2 + y 2 = 0.
∇F = (2x, 2y, 2z), the vector normal of the surface F at P is

1 √ √
n̂1 = [0, (2 17 − 2)1/2 , 17 − 1],
4
∇G = (2x, 2y, −1), the normal vector of the surface G at P is
1 √
n̂2 = √ [0, (2 17 − 2)1/2 , −1].
(2 17 − 1)1/2
∂(F, G) ∂(F, G) ∂(F, G)
= −2y(1+2z), = 2x(1+2z), and = 0. Therefore,
∂(y, z) ∂(z, x) √ ∂(x, y)
at the point P they have the values −[34( 17 − 1)]1/2 , 0 and 0 respectively.
1.2. Parallelism of vector fields 9
The equations of the planes at the point P are according to Eq. (1.14) and
Eq. (1.15)
√ √
(2 17 − 2)1/2 Y + ( 17 − 1)Z = 8,
where Y , Z are coordinates of the plane π1 and

√
√ 1/2 0 0 17 − 1
(2 17 − 2) Y − Z = ,
2
where Y 0 , Z 0 are coordinates of the plane π2 . The equations for the line
lying on both tangent planes which is tangent to the surface at the point
P is given, according to (1.17), by the equations
√ √
X 00 Y 00 − [( 17 − 1)/2]1/2 Z 00 − [( 17 − 1)/2]1/2
√ = = ,
−[34( 17 − 1)]1/2 0 0
or writing these equations in parametric way, it follows

√ √ √
X 00 = −[34( 17−1)]1/2 s, Y 00 = [( 17−1)/2]1/2 , Z 00 = [( 17−1)/2]1/2
(see Fig. 1.9) . The equation of the curve Γ, which is generated by the
intersection both surfaces and passes for the point P , is given by
√
17 − 1
x2 + y 2 = .
2
1.2 Parallelism of vector fields
One of the important mathematical notion for the rest of the book is that
one of parallelism of vector fields.
Definition 1.1. Let a, b be two vector fields continuously defined in region

Ω ⊂ <3 , one says that these vector fields are parallel at P ∈ Ω, P = (x, y, z),
if and only if there is an scalar continuous function λ defined also in Ω such
that
a(P ) = λ(P )b(P ) , (1.20)
and one will say that these vector fields are parallel on Ω if for any P ∈ Ω,
this relation is satisfied. In the case of a and b being constant, λ is just a
real number.
Fig. 1.9 Normal vectors n̂1 and n̂2 to the surfaces S1 and S2 and the tangent line L at
the point P of their interception.
Assume that a = (a1 , a2 , a3 ) and b = (b1 , b2 , b3 ), then, the above rela-

tion can be written as
a1 = λb1 , (1.21a)
a2 = λb2 , (1.21b)
a3 = λb3 , (1.21c)
or in the following way
a1 a2 a3
= = . (1.22)
b1 b2 b3
This last expression is very common and is good to keep it in mind.
Observation 1.1. Multiplying expression (1.21a) by an arbitrary continu-
ous function µ(x, y, z), the expression (1.21b) by another arbitrary contin-
uos function ν(x, y, z) and add them, we would get
µa1 + νa2 = λ(µb1 + νb2 ) (1.23a)
a3 = λb3 , (1.23b)
for any a parallel to b in Ω, which can be written as
µa1 + νa2 a3
= . (1.24)
µb1 + νb2 b3
1.3. Methods of Solution of dx/P = dy/Q = dz/R 11
Observation 1.2. Assume that one of the components of the vector filed
b is zero, says b3 . This means from expression (1.21b) that the component
a3 of the vector field a must be zero, a3 = 0. In terms of the system (3.256),
whenever appears something like
a1 a2 a3
= = , (1.25)
b1 b2 0
this means that a3 = 0 (not singularity of the system).
1.3 Methods of Solution of dx/P = dy/Q = dz/R
We pointed out in the last section that the integral curves of the set of
differential equations
dx dy dz
= = (1.26)
P Q R
form a two-parameter family of curves in three dimensional space. Suppose
we are able to derive from Eq. (1.26) two relations of the form
u1 (x, y, z) = c1 and u2 (x, y, z), (1.27)
where c1 and c2 are the constants of integration, then by varying these
constants, we obtained a two-parameter family of curves satisfying the dif-
ferential equations (1.26).
METHOD (I). Since any tangential direction (dx, dy, dz) at the point
(x, y, z) on the surface u1 (x, y, z) = c1 satisfies the relation
∂u1 ∂u1 ∂u1
dx + dy + dz = 0, (1.28)
∂x ∂y ∂z
and according with the relation (1.19) we also have
∂u1 ∂u1 ∂u1
P+ Q+ R = 0. (1.29)
∂x ∂y ∂z
To find u1 , we look for functions P 0 , Q0 and R0 such that
P 0 P + Q0 Q + Z 0 Z = 0, (1.30)
i.e. a vector field E0 = (P 0 , Q0 , R0 ) which is perpendicular to E = (P, Q, R)
at every point (x, y, z). Because of Eq. (1.29), this vector satisfies
∂u1 ∂u1 ∂u1
P0 = , Q0 = , R0 = . (1.31)
∂x ∂y ∂z
Then, with (1.28) we would have that

P 0 dx + Q0 dy + Z 0 dz
can be an exact differential, du1 (if the integration can be done). The same
procedure can be followed to obtain the other family of curves u2 .
Example 1.2. Find the integral curves of the equation

dx dy dz
= =
x(y − z) y(x − x) z(x − y)
the vector field E is given as
E = (x(y − z), y(x − x), z(x − y)).
If we take the vector fields E0 = (1, 1, 1) and E00 = (zy, zx, xy) the condition
(1.30) is satisfied and the functions u1 , u2 of the equation (1.31) are
u1 = x + y + z, u2 = xyz
hence, the integral curves of the given differential equations are the mem-
bers of the two-parameter family
x + y + z = c1 , xyz = c2 .
We must note that this method depends very much on the intuition and on
the skill to determine the form of the vectors fields E0 , E00 .
METHOD (II). Suppose we find two vectors fields E0 = (P 0 , Q0 , R0 )

and E00 = (P 00 , Q00 , R00 ) such that the differentials
P 0 dx + Q0 dy + Z 0 dz
dw0 = (1.32)
P P 0 + QQ0 + RR0
and
P 00 dx + Q00 dy + Z 00 dz
dw00 = (1.33)
P P 0 + QQ0 + RR0
are exact and are equal to each other. Then it follows
w0 = w00 + c1 (1.34)
where c1 is the integration constant.
Let us choose the vector E0 of the form E0 = (λ, µ, ν), where λ, µ and
ν are constant and find the conditions that they must satisfy in order for
the differential form (1.32), given by
1 λdx + µdy + νdz
, (1.35)
ρ λx + µy + νz
1.3. Methods of Solution of dx/P = dy/Q = dz/R 13
to be an exact differential, where ρ is another constant. From Eq. (ref1.26),

this is possible only if the determinant of the matrix in the expression
    
−ρ b 1 λ 0
 1 −ρ c   µ  =  0  (1.36)
a 1 −ρ ν 0
is zero, that is if ρ is a root of the equation
−ρ3 + (a + b + c)ρ + 1 + abc = 0.
This equation has three complex roots ρi , i = 1, 2, 3 and for each of them
there exists a vector
 
λi
 µi  i = 1, 2, 3
νi
which satisfies Eq. (1.36), and thus we have with Eq. (1.35) three possible
exact differentials
dW 0 = d log(λ1 x + µ1 y + ν1 z)1/ρ1 ,
dW 00 = d log(λ2 x + µ2 y + ν2 z)1/ρ2 ,
and
dW 000 = d log(λ3 x + µ3 y + ν3 z)1/ρ3 .
According to Eq. (1.34), we have the integral curves
(λ1 x + µ1 y + ν1 z)1/ρ1 = c1 (λ2 x + µ2 y + ν2 z)1/ρ2
and
(λ1 x + µ1 y + ν1 z)1/ρ1 = c2 (λ3 x + µ3 y + ν3 z)1/ρ3 .
This method depends also on the intuition in determining the form of the
vector fields E0 , E00 .
METHOD (III). When one of the variables is absent from one the equa-
tions of the set (1.19), it is possible to make a partial separation of vari-
ables, an we can derive the integral curves in a simple way. Suppose that
the equation
dy dz
= ,
Q R
can be written in the form
dy
= f (y, z)
dz
this equation has a solution of the form
φ1 (y, z, c1 ) = 0,
where c1 is the integration constant. Solving this equation for z (z =

ψ(y, c1 )) and substituting this value in the equation
dx dy
= ,
P Q
we obtain an ordinary differential equation of the type
dy
= g(x, y, c1 ),
dx
whose solution
φ2 (x, y, c1 , c2 ) = 0
may be readily obtained.
Example 1.3. Find the integral curves of the equations

dx dy dz
−z
= y = . (1.37)
ye xe y/z
Using the first and third terms, we obtained the ordinary differential equa-
tion
dx
= xe−z ,
dz
which has the solution
−z
x = c1 e−e .
Substituting this in the second and third term of (1.37), we obtain the
ordinary differential equation
dy ey −z
= c21 e−2e .
dz y
This equation has the two parametric solution
Z
−z
(y + 1)e−y + c21 e−2e = c2 .
1.4. Orthogonal Trajectories of a System of Curves on a Surface 15
1.4 Orthogonal Trajectories of a System of Curves on a

Surface
The problem can be set as follows. Given a surface S defined by

F (x, y, z) = 0 (1.38a)
and a system of curves Γ on the surface formed by the set of relations
(1.38a) and
G(x, y, z) = c1 , (1.38b)
find a system of curves {Γ0 } each of which lies on the surface (1.38a) and
cuts every curve of the given system at right angles (see Fig. 1.10). The
new system of curves is called the system of ”orthogonal trajectories” on
the surface of the given system of curves. Because the curves Γ are formed
Fig. 1.10 System of orthogonal trajectories on the sphere an their tangent vectors.
from the intersection of surface (1.38a) and the family of surfaces (1.38b),
the tangential direction of the curves Γ given by (dx, dy, dz) = dξ,~ at the
point, (x, y, z) is perpendicular to the gradients of these surface
∇F · dξ~ = ∇G · ξ~ = 0 (1.39)
at the point, or
∂F ∂F ∂F
dx = dy = dz = 0 (1.40)
∂x ∂y ∂z
and
∂G ∂G ∂G
dx = dy = dz = 0. (1.41)
∂x ∂y ∂z
Hence, the vector dξ~ = (dx, dy, dz) is parallel to the vector ∇F × ∇G (see
Fig. 1.8) and must be such that
dx dy dz
= = , (1.42)
P Q R
where P, Q, and R are defined as
∂(F, G) ∂(F, G) ∂(F, G)
P = ,Q = ,R = . (1.43)
∂(y, z) ∂(z, x) ∂(x, y)
The tangent direction of the curves Γ0 , given by dξ~ = (dx0 , dy 0 , dz 0 ) at the
same point (x, y, z) is perpendicular to the gradient of the surface (1.38a)
and must be perpendicular to the vector field E = (P, Q, R) (i.e ∇F · dξ~0 =
E · dξ~0 = 0), then we have
∂F 0 ∂F 0 ∂F 0
dx = dy = dz = 0 (1.44)
∂x ∂y ∂z
and
P dx0 = Qdy 0 = Rdz 0 = 0. (1.45)
Hence, the vector dξ~0 = (dx0 , dy 0 , dz 0 ) is parallel to the vector ∇F × E and
must be such that
dx0 dy 0 dz 0
= = , (1.46)
P0 Q0 R0
where
∂F ∂F
P0 = R −Q , (1.47a)
∂y ∂z
∂F ∂F
Q0 = P −R , (1.47b)
∂z ∂x
∂F ∂F
R0 = Q −P . (1.47c)
∂x ∂y
The solution of Eq. (1.46) together with the relation (1.38a) give the system
of orthogonal trajectories Γ0 .
Example 1.4. Find the orthogonal trajectories on the sphere x2 +y 2 +z 2 =

a2 of its intersection with the family the of planes
z = k, −a ≤ k ≤ a.
We have the functions
F = x2 + y 2 + z 2 − a 2 = 0
1.5. Pfaffian Differential Equation in R3 17
and
G = z − k = 0,
the functions (1.43) are P = 2y, Q = −2x, and R = 0. The equations for
the curve Γ are
dx dy dz
= = ,
2y −2x 0
which have solutions
x2 + y 2 = a2 − k 2 .
The functions (1.47) are P 0 = −4xz, Q0 = 4yz and R0 = −4(x2 + y 2 ).
The system {Γ0 } of orthogonal trajectories to Γ on the sphere is determined
by the equations
dx dy dz
= =
−4xz −4yz −4(x2 + y 2 )
and
x2 + y 2 = a 2 − z 2 .
The solutions can be written as
xy = c1 , y 2 (x2 + y 2 ) = c2 .
1.5 Pfaffian Differential Equation in R3
The expression
w = P dx + Qdy + Rdz, (1.48)
where (P, Q, R) = E form a vector field in the space, is called a Pfaffian
differential form a 1-form defined in R3 . The equation generated from Eq.
(1.48) (i.e. w = 0) as
P dx + Qdy + Rdz = 0 (1.49)
is called Pfaffian differential equation. This equation has the following
geometric interpretation; let U (x, y, z) = c a family of surfaces orthogonal
to the vector field E and let dξ~ = (dx, dy, dz) be the tangent direction to
a given curve on this surface at the point (x, y, z). The gradient of the
function U (∇U ) is then parallel to the vector field E and thus, equation
(1.49)
E · dξ~ = 0
would represent the family of surfaces perpendicular to the vector field E.

If there exists such a family of surfaces, U (x, y, z) = c, orthogonal to the
vector field E, we say that the Pfaffian differential equation is “integrable”.
If there is not such a family, we say that the Pfaffian differential equation
is “not integrable”. If the Pfaffian is integrable, it means that there exists
a function µ = µ(x, y, z) such that µE = ∇U or
∂U ∂U ∂U
µP = , µQ = , µR = (1.50)
∂x ∂y ∂z
and the function U is obtained through a line integration
Z (x,y,z)
U= µ(P dx + Qdy + Rdz), (1.51)
(x0 ,y0 ,z0 )
if µ = 1, we say that the vector field E is derivable from a “potential

function U ”. We shall see that a necessary condition for the Pfaffian to be
integrable is that E·(∇×E) = 0. In fact, if the Pfaffian (1.49) is integrable,
the relations (1.50) are satisfied and assuming U is twice differentiable with
Uxy = Uyx , Uyz = Uzy , and Uxz = Uzx , we have
∂(µP ) ∂(µQ) ∂(µQ) ∂(µR) ∂(µR) ∂(µP )
= , = , and = (1.52)
∂y ∂x ∂z ∂y ∂x ∂z
and as result of this differentiation, it follows

∂P ∂Q ∂µ ∂Q
µ − =Q −P , (1.53a)
∂y ∂x ∂x ∂y

∂Q ∂R ∂µ ∂Q
µ − =R −Q (1.53b)
∂z ∂y ∂y ∂z
and

∂R ∂P ∂µ ∂Q
µ − =P −R . (1.53c)
∂x ∂z ∂z ∂x
Multiplying Eq. (1.53a) by R, Eq.(1.53b) by P , Eq. (1.53c) by Q, and
adding them, we get

∂P ∂Q ∂Q ∂R ∂R ∂P
R − +P − +Q − =0 (1.54)
∂y ∂x ∂z ∂y ∂x ∂z
which is precisely the expressions E · (∇ × E) = 0. It happens that is
expressions is sufficient to assure that the Pfaffian (1.49) is integrable. We
will not demonstrate this fact here. Once we know that the Pfaffian is
integrable, we can use the relation (1.50) to find the family of surfaces
orthogonal to the vector field E.
1.6. N. Mechanics, Lagrangians, Hamiltonian, H-J Equation, and Liouville’s Theorem19
Example 1.5. Verify that the differential equation

ydz + xdy + 2zdz = 0 (1.55)
is integrable and find its primitive. In this case
E = (y, x, 2z) (1.56)
so that ∇ × E = (0, 0, 0) and it is clear that E · (∇ × E) = 0 and then µ = 1.
From Eq. (1.50) and Eq. (1.56), we have
∂U ∂U ∂U
y= , x= , and 2z =
∂x ∂y ∂z
upon integration, the solution is
U = xy + z 2 . (1.57)
An alternative way to find the solution is to notice from (1.55) that it can
be expressed as the exact differential
d(xy + z 2 ) = 0
so that solution (1.57) is again obtained.
1.6 Newton’s Mechanics, Lagrangians, Hamiltonian,

Hamilton-Jacobi Equation, and Liouville’s Theorem
N -dimensional Newton’s mechanics can be seen as a particular dynamical

system described by the set the equations
dvi
= fi (q, v, t), i = 1, ..., N (1.58a)
dt
and
dqi
= vi , i = i, ..., N, (1.58b)
dt
where q = (q1 , ..., qN ), v = (v1 , ..., vN ) and “t00 represented the generalized
coordinates, generalized components of the velocity, and the time parameter
respectively. fi represents the total force acting on the particle in the
direction “i00 divided by the constant mass of the particle. Eq. (1.58a) and
(1.58b) describe the motion of a single particle moving in a N -dimensional
space under given force. These equations can be written in the same form
as the relation (1.19)
dq1 dqN dv1 dvN
= ... = = = ... = = dt. (1.59a)
dv1 dvN df1 dfN
The solution equations (1.58) or (1.59a) represent curves (or trajectories)

in a 2N -dimensional space (q, v) whose tangent directions are parallel to
the vector field
E = (v1 , ...vN , f1 , ...fN ). (1.59b)
For most of the physical problems, it is possible to deduce Eq. (1.58a) and
Eq. (1.58b) from a variational principle (see reference of J. Douglas for the
case N = 2),
Z b
δ Ldt = 0, (1.60)
a
where L is called the Lagrangian of the system (1.58) and is a function
depending on the generalized coordinates, velocities, and time
L = L(q, q̇, t). (1.61)
where q̇i can be the ith-component of the velocity, q̇i = vi . As a result of
the variation of Eq. (1.60), we obtained the Euler equations (see reference
of H. Goldstein for more details)

d ∂L ∂L
− = 0, i = 1, ..., N. (1.62)
dt ∂ q̇i ∂qi
The usual approach in physical problems is starting with a more or less un-
derstood Lagrangian (1.61) to deduce the equations (1.58) through its dif-
ferentiation (1.62). The inverse problem, however, starts with Eq. (1.58a)
and Eq. (1.58b) to deduce the Lagrangian (1.61), this procedure is not so
simple and is called “The Inverse Problem of the Calculus of Variations”.
Once we have the Lagrangian (1.61), we define a new variable pi (called
the generalized linear momentum) as
∂L
pi = (1.63)
∂ q̇i
and make a Legendre Transformation,
N
X
H= q̇i pi − L, (1.64)
i=1
to define the Hamiltonian, H = H(q, p, t), of the system (1.58). Through
this procedure, it is assumed that we can have q̇ = q̇(q, p, t) to be able to
substitute this on the right hand side of Eq. (1.64). Using Eq. (1.63) and
Eq. (1.62), we get
N N
X ∂H ∂H ∂H X ∂L
dH = dqi + dpi + dt = (q̇i dpi − ṗi dqi ) −
i=1
∂q i ∂pi ∂t i=1
∂t
Equaling coefficients and using the fact that the variables qi and pi are
independent i = 1, ..., N , we obtain the Hamilton equations of motion
∂qi ∂H
= , i = 1, ..., N (1.65a)
∂t ∂pi
∂pi ∂H
=− , i = 1, ..., N (1.65b)
∂t ∂qi
and
∂L ∂H
=− . (1.65c)
∂t ∂t
Note that Eq. (1.65a) and Eq. (1.65b) can be written in the form (1.19) as
dq1 dq dp dp
= ... = N = 1 = ... = N . (1.66)
∂H ∂H ∂H ∂H
− −
∂p1 ∂pN ∂q1 ∂qN
In this way, the motion of the particle in N -dimensions can be described
by equations (1.58) or by equations (1.65). However, Eq. (1.65a) and
Eq. (1.65b) give us certain advantages to observe the dynamics of the
system. The solutions of equation (1.65) or (1.66) given us curves in a 2N -
dimensional space (q, p) called “Phase Space” whose tangent directions are
parallel to the vector field

∂H ∂H ∂H ∂H
E= , ..., ,− , ..., − .
∂p1 ∂pN ∂q1 ∂qN
We must notice that the existence of the Hamiltonian for system (1.58)
depends on the existence of the Lagrangian, through the Legendre trans-
formation (1.64) (see reference of D. Darboux for N = 1 and the reference of
J. Douglas for N = 2), and given the Hamiltonian, the problem of the direct
determination of the Lagrangian may be made using Eq. (1.63) and Eq.
(1.64) through the solution of the nonlinear (in general) partial differential
equation of first order
X N
∂L ∂L
H q, ,t − q̇ + L = 0. (1.67)
∂ q̇ i=1
∂ q̇
This expression represents a non linear partial differential equation of first

order for the function L. Now, because the variables qi and pi i = 1, ...N
are independent, we can do, in general, a change in variable in the phase
space of the form
Qi = Qi (q, p, t), Pi = Pi (q, p, t) i = 1, ..., N (1.68)
satisfying, of course, the condition that the Jacobian of the transformation

be different from zero,
∂(Q, P )
6= 0
∂(q, p)
(if Qi = Qi (q, t) and Pi = Pi (q, t), i = 1, ..., N the transformation is called
a point transformation). If under transformation (1.68), we have a function
Ve = Ve (Q, P, t) such that
∂ Ve ∂ Ve ∂Le ∂ Ve
Q̇i = , Ṗi = i = 1, ..., N and =− , (1.69)
∂Pi ∂Qi ∂t ∂t
that is, the Hamilton equations (1.65) remain invariant, then the trans-
formation (1.68) is called “canonical”. It can be demonstrated that this
happens when the Jacobian of the transformation is equal to 1, that is, the
volume of the phase space is conserved.
Suppose that the Eq. (1.68) represents a canonical transformation.

Since it satisfies Eq. (1.69), we can form the Lagrangian L
e = L(Q,
e Q̇, t)
with a Legendre transformation like Eq. (1.64)
N
X
L
e= Q̇i Pi − Ve (1.70)
i=1
which will satisfy a similar principle to Eq. (1.60), given as
Z b
δ Ldt
e = 0, (1.71)
a
but since both of them describe te same system (1.58), they must differ a
total time derivative of a function F , that is
L=L e + dF . (1.72)
dt
This function F is called the “generating function” and once this one is
given, the transformations (1.68) are completely specified. F makes the
connection of the old variables (q, p) and the new ones (Q, P), thus F
is a function of the old and new variables. There are four possibilities
F1 (q, Q, t), S(q, P, t), F3 (p, Q, t), and F4 (p, P, t). Let us consider the sec-
ond possibility, from Eq. (1.64), Eq. (1.70) and Eq. (1.72), using the fact
PN
S(q, P, t) = F1 (q, Q, t) + i=1 Pi Qi and taking the total time differentia-
tion of F1 , we obtain
N N
X ∂S X ∂S ∂S
pi − q̇i + Qi − Ṗi + Ve − H + = 0,
i=1
∂q i i=1
∂Pi ∂t
and, therefore, we get the equations

∂S
pi = , i = 1, ..., N, (1.73a)
∂qi
∂S
Qi = , i = 1, ..., N (1.73b)
∂Pi
and
∂S
Ve = H + . (1.74)
∂t
Let us assume now that the transformation generated is such that Q̇i = 0
and Ṗi = 0, i = 1, ..., N which implies
Qi = βi , Pi = αi (constants) i = 1, ..., N (1.75)
and from Eq. (1.69), we can make Ve ≡ 0, then from Eq. (1.74), we have

∂S ∂S
H q, ,t + = 0. (1.76)
∂q ∂t
Because of the relation (1.75), we have
S = S(q, α, t) (1.77)
and
∂S
βi = i = 1, ..., N. (1.78)
∂αi
Eq. (1.76) is called “The Hamilton-Jacobi equation” associated to the sys-
tem (1.58). Solving this equation and using Eq. (1.73a) and (1.73b), we can
solve our physical problem. S is called “The Hamilton principal function”
whose existence depends on the Hamiltonian through the relation (1.74).
This expression represents a non linear partial differential equation of first
order for the function S(q, α, t).
In this way there are three possible ways to solve a physical problem
in Classical Mechanics using Eq. (1.58) (Newton’s equations), Eq. (1.65)
(Hamilton’s equations) or Eq. (1.76) (Hamilton-Jacobi’s equation). Con-
sider now a system of N particles moving in a three dimensional space, then
as a first approach we could use any of the above three ways to describe our
system, but it is almost impossible to be able to do this with the method
mentioned before (so called many bodies problem). An alternative way is
to use statical mechanics where “the state of a particle” is represented as a
point in the 6N -dimensional phase space (q, p), and “the state of the whole
system” is represented by density function ρ(q, p, t) (number of particles
per unit phase space volume, dN e /dV ). This density is called “ensemble”
or “representative set” of the system. To see the motion of ρ in the phase
space we take the total time derivative of ρ
3N
dρ X ∂ρ ∂ρ ∂ρ
= q̇i + ṗi + .
dt i=1
∂qi ∂pi ∂t
Using Eqs. (1.65), we have

3N
dρ X ∂ρ ∂H ∂ρ ∂H ∂ρ
= − + . (1.79)
dt i=1
∂qi ∂pi ∂pi ∂qi ∂t
Defining the poisson bracket of two functions f = f (q, p, t) and g =

g(q, p, t) as
3N
X ∂f ∂g ∂f ∂g
{f, g} = − , (1.80)
i=1
∂qi ∂pi ∂pi ∂qi
Eq. (1.79) can be written as

dρ ∂ρ
= {ρ, H} + . (1.81)
dt ∂t
If the number of points a fixed volume is constant (there is not annihi-
lation or creation of points, sinks our sources) and the volume remains
constant for a fixed number of particles (Liouville’s Theorem), for a system
in thermodynamic equilibrium where ρ does not depend explicitly on time
(∂ρ/∂t = 0), we have the equation
{ρ, H} = 0. (1.82)
For a system out of equilibrium, we have ∂ρ/∂t 6= 0.
Given the Hamiltonian for the N -particle system in three dimensions,

Eq. (1.81) or Eq. (1.82), represents a partial differential equation of first
order for ρ.
Example 1.6. The Lagrangian for one a dimensional single particle in a

dissipative medium, where the friction force is proportional to the speed
squared, is given by
1 2
L= mq̇ exp (2αq/m), (1.83)
2
where m is the mass of the particle and α is a friction coefficient. Find the
equation of the motion in the form (1.58) and (1.65), and give the partial
differential equations Eq. (1.67), Eq. (1.76), and Eq. (1.82).
Using the Euler’s equations (1.62) for N = 1, we have

mq̈ = −αq̇.
Then, the equations the motion in the form (1.58) are
dv α
= − v2 , (1.84a)
dt m
dq
= v. (1.84b)
dt
Using Eq. (1.63), the generalized linear momentum is
p = mq̇ exp (2αq/m), (1.85)
and substituting this in the Legendre transformation, Eq. (1.64), the
Hamiltonian is given by
p2
H= exp (−2αq/m). (1.86)
2m
Thus, Hamilton’s equations of motion are
dq p
= exp (−2αq/m) (1.87a)
dt m
and
dp αp2
= 2 exp (−2αq/m). (1.87b)
dt m
With Eq. (1.86) and Eq. (1.76), the Hamilton-Jacobi equation is given by
2
1 ∂S ∂S
exp (−2αq/m) + = 0, (1.88)
2m ∂q ∂t
and using Eq. (1.82), the Liouville’s theorem is expressed as
∂ρ αp ∂ρ ∂ρ
+ + = 0. (1.89)
∂q m ∂p ∂t
Finally, given the Hamiltonian (1.86), one way to find the Lagrangian (1.83)
directly is through the solution of Eq. (1.67), i.e.
2
exp (−2αq/m) ∂L ∂L
− q̇ + L = 0, (1.90)
2m ∂ q̇ ∂ q̇
and it is not so difficult to demonstrate that the Lagrangian (1.83) indeed
satisfies Eq. (1.90).
Example 1.7. The dynamical equations of a physical pendulum of mass

“m”, length “l”, and quadratic velocity dissipation can be written as
dθ
=v (1.91a)
dt
and
dv α
= −ω02 sin θ − v|v|. (1.91b)
dt m
The function K = (θ, v) given by
2ω02

2αsig(v) m exp (2αθsig(v)/m)
K = v2 + sin θ − cos θ
1 + (2α/m)2 m 2
(1.92)
is a constant of motion since it satisfies the partial differential equation
∂K 2 α ∂K
v − ω0 sin θ + v|v| = 0. (1.93)
∂θ m ∂v
The function sig(v) and the constant ω0 are defined as

+1, if v > 0
sig(v) =
−1, if v < 0
and
p
ω0 = l/g,
being g the local acceleration due to gravity. Properly, (1.92) is a constant of

motion restricted for v > 0 or v < 0 since whenever there is a cross through
v = 0, its value must be changed to bring about the spiral shrinking motion
of the particle in the phase space. The Lagrangian of the system is given
by
2ω02 m exp ( 2αθsig(v)

2 2αsig(v) m )
L(θ, v) = v − 2
sin θ − cos θ
1 + (2α/m) m 2
(1.94)
and generalized linear momentum by
p(θ, v) = mv exp (2αθsig(v)/m). (1.95)
From this expression, we get

p
v(θ, p) = exp (−2αθsig(v)/m). (1.96)
m
and the Hamiltonian of the system is given by
p2
H(θ, p) = exp (−2αθsig(v)/m)
2m
mω02 exp (2αθsig(v)/m) 2αsig(v)

+ sin θ − cos θ = 0. (1.97)
1 + (2α/m)2 m
Then, the Hamilton-Jacobi equation associated to this system is written as
2
mω02 exp (2αθsig(v)/m) 2αsig(v)

1 ∂S ∂S
+ + sin θ − cos θ = 0.
2m ∂θ ∂t 1 + (2α/m)2 m
(1.98)
1.6.1 Equivalent Hamiltonians

As we saw in section (1.6), two Lagrangians L(q, q̇, t) and L(Q,e Q̇, t) are
equivalents if there is a function F1 , called generatrix function, such that
L=L e + dF1 /dt. This generatrix function is of the form F1 (q, Q, t) and is
related with the other possible generatrix functions, F2 (q, P, t), F3 (p, Q, t),
and F4 (p, P, t) through a Legendre transformation. To give this generatrix
function is the same as to give the canonical transformation which changes
the Hamiltonian ( Lagrangian) into another one. The transformed Hamil-
tonian, H,
e is related with the initial Hamiltonian, H, in the form
He = H + dFi , (1.99)
dt
independently on which generatrix function is used, i = 1, 2, 3, 4. Therefore,
we can say that two Hamiltonian H(q, p, t) and H(Q, e P, t) are equivalents
if the above relation is satisfied for any generatrix function. Note that the
units of both Hamiltonians must be the same to be able to have this ex-
pression. Given the generatrix function Fi , there are differential relations
with its complementary variables, for example, for F1 (q, Q, t) its comple-
mentary variables are related by pj = ∂F1 /∂qj and Pj = −∂F1 /∂Qj . So,
the equivalence of the Hamiltonians brings about the following non-linear
partial differential equations
∂F1
H(Q,
e −∇Q F1 , t) = H(q, ∇q F1 , t) + , (1.100)
∂t
∂F2
H(∇
e
P
F2 , P, t) = H(q, ∇q F2 , t) + , (1.101)
∂t
∂F3
H(Q,
e −∇Q F3 , t) = H(−∇p F3 , p, t) + , (1.102)
∂t
∂F4
H(∇
e
P
F4 , P, t) = H(−∇p F4 , p, t) + , (1.103)
∂t
where ∇ξ represents the gradient with respect the variable ξ. Because of
the structure of these partial differential equations of first order, there may
be a solution of separable variables (depending of the Hamiltonian expres-
sions) , for example F1 (q, Q, t) = f1 (q, t) + g(Q, t). However, this type of
solution defines only a point-like transformation. Therefore, it is necessary
to look for different type of solution.
Example 1.8. We saw already on previous section that any Hamiltonian

H(q, p, t) is equivalent to the Hamiltonian H e = 0 if we choose the gener-
atrix function F2 (q, P, t) = S(q, t) as the Hamilton-Jacobi function, with
P=α ~ (constant). Then, this generatrix function satisfies the Hamilton-
Jacobi equation (Eq. 1.76 above)
∂S
H(q, ∇q S, t) + =0. (1.104)
∂t
Example 1.9. Write the possible equations that establish the equivalence
between the harmonic oscillator, H = p2 /2m + mω 2 q 2 /2, and the Hamil-
e = ωP with q, p, P ∈ R.
tonian H
From the above equation one has
2
∂F1 1 ∂F1 1 ∂F1
−ω = + mω 2 q 2 + (1.105)
∂Q 2m ∂q 2 ∂t
2
| ∂F2 1 ∂F2
ωP = + mω 2 q 2 + (1.106)
2m ∂q 2 ∂t
2
p2

∂F3 1 2 ∂F3 ∂F3
−ω = + mω + (1.107)
∂Q 2m 2 ∂p ∂t
2
p2

1 ∂F4 ∂F4
ωP = + mω 2 + . (1.108)
2m 2 ∂p ∂t
which defines non linear partial differential equations in R3 .
Of course, once we know a generatrix function, the other ones are de-
termined through a Legendre transformation, for example, assume that we
have determine the generatrix function F1 (q, Q, t), then, since the gener-
atrix function F2 (q, P, t) is related with F1 through the Legendre trans-
formation F1 (q, Q, t) = F2 (q, P, t) − Q · P, and since one has the relation
Qi = ∂F2 /∂Pi , we cant think this Legendre transformation as the following

non linear partial differential equation of first order
X ∂F2
Pi + F1 (q, ∇P F2 , t) = F2 (q, P, t) (1.109)
i
∂Pi
In the same way, assuming F1 (q, Q, t) is known, one could have the gen-
eratrix functions F3 and F4 determined by the following non linear partial
differential equations of first order
X ∂F3
pi + F1 (−∇p F3 , Q, t) = F3 (p, Q), t) (1.110)
i
∂pi
X ∂F4 ∂F4

pi + Pi + F1 (−∇p F4 , ∇P F4 , t) = F4 (p, P, t) (1.111)
i
∂pi ∂Pi
Example 1.10. Let us assume that the generatrix function F1 is given as

mω 2 q 2
F1 (q, Q) = − tan Q, (1.112)
2
where m and ω are real parameters. This generatrix function establishes
the equivalence between the Hamiltonian H = p2 /2m + mω 2 x2 /2 and the
Hamiltonian H e = ωP since is a non separable solution of the equation
−ω(∂F1 /∂Q) = (∂F1 /∂q)2 /2m + mω 2 q 2 /2. Then, according to what we
said above, the generatrix function F2 (q, P ) can be found as the solution
of the non linear partial differential equation
mω 2 q 2

∂F2 ∂F2
P − tan = F2 (q, P ) • (1.113)
∂P 2 ∂P
Note from the Legendre transformation that there is always a variable which
is not involved in the differentiation, in our last example this variable is
”q.” This fact also happens for the Legendre transformation between the
Lagrangian, L(q, q̇, t), and the Hamiltonian, H(q, p, t),
X ∂L
q̇i − H(q, ∇q̇ L, t) = L(q, q̇, t) (1.114)
i
∂ q̇i
In fact, if this Legendre transformation is though as a function of q and
q̇ instead, and by calling K(q, q̇, t) = H q, p(q, q̇, t), t , one would have
instead the following linear partial differential equation of first order
X ∂L
q̇i = L(q, q̇, t) + K(q, q̇, t) (1.115)
i
∂ q̇
1.7 Problems
√
1.1 Find the tangent planes at the point P = (0, 3, 1) of the surfaces
x2 + y 2 + z 2 = 4 and z = 1. Find the normal vectors to these tangent
planes at the point, the line generated by intersection of these planes, and
the curve generated by the intersection of both surfaces.
1.2 Find the integral curves of the follows equations, using the method
(I) of the sec.1.3
dx dy dz
= = .
xz − y yz − x 1 − z2
1.3 Find the integral curves of the equations, using the method (II) of
the sec.1.3
dx dy dz
= = .
y + az z + bx x + cy
1.4 Find the integral curves of the equations, using the method (II) of
the sec.1.3
adx bdy cdz
= = .
(b − c)yz (c − a)zx (a − b)xy
1.5 Find the integral curves of the equations,using the method (III) of
the sec.1.3
dx dy dz
= = .
x+z y z + y2
1.6 Find the orthogonal trajectories on the cone x2 + y 2 = z 2 tan2 α of
its intersection with the family of planes parallel to z = 0.
1.7 Verify that the differential equation

(y 2 + yz)dz + (xz + z 2 )dy + (y 2 − xy)dz = 0
is integrable and find its primitive.
1.8 Do the same as on the Exercise 6, but for the one dimensional oscil-
lator whose Lagrangian is given by
1 1
L = mq̇ 2 − kq 2 ,
2 2
where m is the mass of the particle and k is the spring constant.
1.9 Write down the non-linear partial differential equation to find the
Lagrangian directly if the Hamiltonian is given by (1.97).
1.7. Problems 31
1.10 Write the equation which would establish the equivalence between
the Hamiltonians H1 = p2 /2m + k/q and H2 = p2 /2m + mω 2 q 2 /2 (select
one Hamiltonian with variable Q and P ).
P2
1.11 Given the generatrix function F1 = i=1 (tan Qi − β̇i (t))x2i /2βi (t),
write down the partial differential equation of first order which helps to find
the generatrix function F2 (x1 , x2 , P1 , P2 , t).
1.12 Given the Euler-Lagrange equation for one degreed of freedom,

dLv
− Lx = 0 ,
dt
where one has defined Lv = ∂L/∂v and Lx = ∂L/∂x, make the differenti-
ation of this expression with respect to ”v,” and show that a PDEFO with
respect to J = Lvv given by
∂J ∂J
+ F (x, v)
v = −Fv J ,
∂x ∂v
where dx/dt = v and dv/dt = F (x, v) is the defined autonomous system.
1.13 If an arbitrary conservative system with three degrees of freedom is

characterized by the potential function V (x), show (using Euler-Lagrange
equations) that

1
L= mv − V (x) eαt/m
2
2
represents an explicitly time depending Lagrangian for the associated dis-
sipative system
dv
m = F(x) − αv ,
dt
where v is the velocity of the particle with speed v and constant mass m,
F(x) = −∇V (x) is the conservative force, and α is a non negative real
constant.
3.14 Show that the Hamiltonian and Hamiton equations for the previous
exercise are given by
p2 −αt/m pi −αt/m ∂V αt/m
H= e + V (x)eαt/m , ẋi = e , e ṗi = −
,
2m m ∂xi
(1.116)
and write the Hamilton-Jacobi equation associated to this problem.
1.8 References
1.1 I.N. Sneddon, Elements of Partial Differential Equations, Mc Graw

Hill Books Company, Inc. 1957. Chap.I
1.2 F.H. Miller, Partial Differential Equations, John Wiley & Sons, Inc.
1941. Chap. II.
1.3 L. Elsgoltz, Ecuaciones Diferenciales y Cálculo Variacional, Ciencia

1975, Chap. 5.
1.4 M. Spivak, Calculus on Manifolds, W.A. Benjamin, Inc., 1965. Chap.

4.
1.5 H. Goldstein, Classical Mechanics, Addison-Wesley Publishing Co.

Inc., 1965. Chap.1.
1.6 J. Douglas, Trans. Amer. Math. Soc. 50, (1947) 71.
1.7 I.M. Gelfand and S.V. Fomins, Calculus of Variations, Prentice-Hall,

Inc. ,1963. Chap. 1-5.
1.8 G. López, Ann. Phys., 251, No.2 (1996) 372.
1.9 J.A. Kobussen, Acta Phys. Austr. 51 (1979) 293.
1.10 C. Leubner, Phys. Lett. A, 86 (1981) 2.
1.11 D. Darboux, Leçons sur la théorie général des surfaces et les appli-
cations géométriques du calcul infinetésimal, Gauthier-Villars, Paris, 1984.
IViéme partie,
Chapter 2
Partial Differential Equations of First

Order
In this chapter we make the introduction to the Partial Differential Equa-

tion of First Order (henceforth abbreviated as PDEFO) considering the
classification and the solution of a particular kind of PDEFO.
2.1 Classification
A partial differential equation for a function z defined in a n-dimensional

space is an arbitrary relation of the form
F (x, z, zi ; zi1 i2 ; ...; zi1 ,...,im ) = 0 (2.1)

where x = (x1 , ..., xn ) is a point in the space Rn , zi , zi1 i2 and zi1 ,...,im are
the partial derivatives defined as
∂z
zi = i = 1, ..., n, (2.2a)
∂xi
∂2z
zi1 i2 = i1 , i2 = 1, ..., n, (2.2b)
∂xi1 ∂xi2
and
∂mz
zi1 ,...,im = ij = 1, . . . , n (2.2c)
∂xi1 , ..., ∂xim
where m is a natural number. We call z = φ(x1 , ..., xn ) a solution of (2.1) if
after substitution of z and its partial derivatives, Eq. (2.1) is satisfied iden-
tically in every point x of some region Ω ⊂ Rn of the space. We will assume
also that the function z and its derivatives (2.2) up to the order k ∈ Z are
continuous in the region Ω that is, z is a class C k (Ω), where C k (Ω) denotes
the set of real functions defined in Ω with its derivatives continuous up to
33
34 Partial Differential Equations of First Order
the order k. It is not difficult to prove that C k (Ω) forms a vector space
over the real space R. The order of the partial differential equation is the
order of the highest derivative that occurs in (2.1), the natural number m.
A PDEFO for a function z defined in region Ω of the space Rn , is a

relation of the form

∂z ∂z
F x, z, , ..., = g(x, z), (2.3)
∂x1 ∂xn
where the pure dependence on the x and z has been separated through
the function g(x, z). A solution z = f (x1 , ..., xn ) of Eq. (2.3), when inter-
preted as a n-dimensional surface, will be called an “integral surface” of
the differential equation.
The PDEFO (2.3) can be see, fixing the point x ∈ Ω, as a functional F
acting in the space C k (Ω) with values in the real space R. According
with this observation, we will say that a PDEFO is linear if for every
z1 , z2 ∈ C k (Ω) and α ∈ R, we have

∂(z1 + αz2 ) ∂z1 ∂z2
F x, z1 + αz2 , = F x, z1 , + αF x, z2 , (2.4)
∂xi ∂xi ∂xi
i.e., Eq. (2.3) is linear with respect to z and its partial derivatives. We will
say that a PDEFO is quasi-linear if instead of the relation (2.4), we have

∂(z1 + αz2 ) ∂z1
F x, z1 + αz2 , = F1 x, z1 + αz2 ,
∂xi ∂xi

∂z2
+ αF2 x, z1 + αz2 , , (2.5)
∂xi
where F1 and F2 could be different functions. That is, Eq. (2.3) is only
linear with respect to the partial derivatives. We will say that the PDEFO
is non-linear if neither relation (2.4) or (2.5) is satisfied.
A general linear PDEFO can be written as
n
X ∂z
ai (x) − c(x)z − d(x) = 0, (2.6)
i=1
∂xi
and a general quasi-linear PDEFO can be written as

2.2. Linear PDEFO for Functions Defined in Ω ⊂ R2 35
n
X ∂z
ai (x, z) − c(x, z) = 0. (2.7)
i=1
∂xi
An example of a non-linear PDEFO is the following equation
n 2
X ∂z
ai (x) =0. (2.8)
i=1
∂xi
2.2 Linear PDEFO for Functions Defined in Ω ⊂ R2
The theory we will discuss in this section is extended immediately to any

number of variables (dimensions). We restrict ourselves to two variables
(x, y) ∈ Ω ⊂ R2 to make clear the interesting geometric interpretation of
the linear PDEFO. Eq. (2.6) can be written as
∂z ∂z
a(x, y) + b(x, y) = c(x, y)z + d(x, y) (2.9)
∂x ∂y
We notice that the left hand side of this equation represents the deriva-
tive of z(x, y) in the direction
E(x, y) = (a(x, y), b(x, y))

that is,
∂z ∂z
a(x, y) + b(x, y) = E · ∇z
∂x ∂y
thus, when we consider the curves in the x − y plane whose tangents (dξ)~
at each point have those directions (dξ~ ∼ E); that is, the one parametric
family of curves defined by the ordinary differential equations
dx dy
= a(x, y), = b(x, y) (2.10a)
dt dt
or the single equation
dx a(x, y)
= , (2.10b)
dy b(x, y)
is such that along this curve, z(x, y) will satisfy the ordinary differential
equation
dz
= c(x, y)z + d(x, y) (2.11a)
dx
or
dz c(x, y)z + d(x, y)
= . (2.11b)
dx a(x, y)
Let us prove this. Differentiating z with respect to the parameter t, we
have
dz ∂z dx ∂z dy
= + ,
dt ∂x dt ∂y dt
along the curves (2.10), this expression becomes
dz ∂z ∂z
= a(x, y) + b(x, y) ,
dt ∂x ∂y
but according to Eq. (2.9), we get Eq. (2.11). The one parametric family
of curves defined by Eq. (2.10) are called the “characteristic curves” of the
differential equation, which will be denoted by {cλ }.
Suppose now that z(x, y) is assigned an “initial” value at the point x0 , y0

in the x − y plane. From the existence and uniqueness of the initial value
problem for ordinary differential equations, Eq. (2.10) will define a unique
characteristic curve, say
x = x(x0 , y0 , t), y = y(x0 , y0 , t) (2.12)

along which
z = z(z0 , x0 , y0 , t) (2.13)
will be uniquely determined by Eq. (2.11). That is, If z is given at a point,
it is determined along a whole characteristic curve through the point. This
suggests that if we were to assign initial values for z along some curve Γ
(Fig. 2.1), intersecting the characteristic cλ , we may determine a unique
solution z(x, y) in the whole region covered by the family cλ by means of
Eq. (2.12) and Eq. (2.13).
We can anticipate problems with the determination of the solution z(x, y)

2.2. Linear PDEFO for Functions Defined in Ω ⊂ R2 37
Fig. 2.1 The initial curve Γ determine an unique solution z(x, y), meanwhile z(x, y) is
not unique for Γ0 .
for curves like Γ0 shown in Fig. 2.1 because for several points of the char-
acteristic curve c0 , we have the same solution (2.13). The curve Γ, which
we may call the initial curve, may not be chosen quite arbitrary. Clearly, it
must not coincide with the characteristic curve since z must be determined
as a solution of an ordinary differential equation in unique form.
A precise formulation of this initial value problem, called the Cauchy

initial value problem, will be given for the more general quasi-linear equa-
tions to follow. However, the next example points out some important
feature.
Example 1. Find the integral surface of the equation
e−x

1 ∂z ∂z
+ = z
y ∂x ∂y x
with initial conditions z = φ(x) for y = 1. The equation for the character-
istic curve is given by
dy
=y
dx
having the solutions
y = cex .
Along such a curve, z satisfies Eq. (2.11b)
dz ye−x c
= z= z
dx x x
z = K(c)xc .
The constant K depends on c because it may differ from characteristic
to characteristic, then we have the general solution
−x
z(x, y) = K(ye−x )xye , (2.14)
where K is “arbitrary function”. If we apply the initial condition for y = 1,
we obtain
−x
φ(x) = K(e−x )xe
or
φ(log s−1 )
K(s) = ,
(log s−1 )s
hence, the functionality K(ye−x ) is determined, and the required solution
is
−x
φ(x − log y)xye
z(x, y) = .
(x − log y)ye−x
Let us see what happens if instead of having the above initial curve
y = 1, we choose the characteristic curve given by y = 2ex . With the initial
condition z = φ(x) at y = 2ex substituted in (2.14), we can not determine
the functionality of K because we get the following relation
φ(x) = K(2)x2 .
Thus, the Cauchy problem is not well posed on characteristic curves. We
need to take care that our initial curve be different from any characteristic
curve of the differential equation.
2.3. Quasi-linear PDEFO for Functions Defined in Ω ⊂ R2 39
2.3 Quasi-linear PDEFO for Functions Defined in Ω ⊂ R2
According with (2.7), the general quasi-linear equation may be written as

∂z ∂z
P (x, y, z) + Q(x, y, z) = R(x, y, z). (2.15)
∂x ∂y
The solution z = f (x, y) defines an integral surface in the three dimen-

sional x, y, z space. As we saw in the relation (1.38), the directions numbers
of the normal to the surface are (∂z/∂x, ∂z/∂y, −1), so that equation (2.15)
can be interpreted as the condition for the integral surface at each point to
have the property that the vector field E = (P, Q, R) is tangent to the sur-
face. Thus, the quasi-linear equation (2.15) defines a direction field E called
the characteristic direction, having the property that a surface z = f (x, y)
is an integral surface if and only if at each point(x, y, z) the tangent plane
contains the characteristic direction (see Fig. 2.2).
Fig. 2.2 For a surface z = f (x, y) the field E gives us the characteristic directions.
It is suggestive then that we consider the integral curves of this field E

i.e., the family of space curves whose tangent direction dξ~ = (dx, dy, dz)
coincides with the direction of the vector field E (dξ~ ∼ E). They are called
the characteristic curves and are given by equations

dx dy dz
= = . (2.16a)
P (x, y, z) Q(x, y, z) R(x, y, z)
Calling the common value of these ratios dt (the constant of proportionality
between the vector dξ~ and E). We can write Eq. (2.16a) in the form
dx dy dz
= P (x, y, z), = Q(x, y, z), = R(x, y, z) (2.16b)
dt dt dt
This notion differs from the one used in the linear case. The projection
of the present curves on the x − y plane will be the curves previously called
the characteristics. Through each point (x0 , y0 , z0 ) passes one characteristic
curve
x = x(x0 , y0 , z0 , t), y = y(x0 , y0 , z0 , t), z = z(x0 , y0 , z0 , t).
One important property of the characteristic curves is immediately ev-
ident from the geometric interpretation of equation (2.15). Namely, every
surface generated by one parametric family of characteristic is an integral
surface. Moreover, the converse is also true, if z = f (x, y) is a given integral
surface S. Consider the solution of
dx dy
= P (x, y, f (x, y)), = P (x, y, f (x, y))
dt dt
with x = x0 , y = y0 for t = 0. Then, for the corresponding curve
x = x(t), y = y(t), z = f (x(t), y(t)),
we have from Eq. (2.15)
dz dx dy
= fx + fy = P (x, y, z)fx + Q(x, y, z)fy = R(x, y, f ).
dt dt dt
Hence, the curve satisfies condition (2.16b) for characteristic curves, and
also lies on S by definition. Thus, S contains together with each point
also the characteristic curve through the point. Therefore, if two integral
surfaces intersect at a point, they intersect along the whole characteristic
through this point because of the uniqueness of the solution (see below);
and the curve of intersection of two integral surfaces must be characteristic.
The integral surface is also called the vector surface because the tangent
of the curves lying on it have the same direction of the vector field E. If
the integral surface is given implicitly of the form
F (x, y, z) = 0. (2.17)
The normal vector n̂ at any point of the surface is given by the normal-
ization to unit of the vector

∂F ∂F ∂F
∇F = , , (2.18)
∂x ∂y ∂z
and the integral surface is determined by the condition of the vector field
E which is orthogonal to the normal vector of the surface (2.17)
∂F ∂F ∂F
E · ∇F = P (x, y, z) + Q(x, y, z) + R(x, y, z) = 0. (2.19)
∂x ∂y ∂z
The characteristic curves of Eq. (2.19) are determined by the same set of
equations (2.16). We must note that Eq. (2.19) is a linear PDEFO with
three variables (x, y, z).
Let us assume that we have two independent characteristic curves when

solving Eq. (2.16a), given as
ψ1 (x, y, z) = c1 , ψ2 (x, y, z) = c2 . (2.20)
These are solution of Eq. (2.19). To show this fact, let us take the
differential of one of the solutions given by Eq. (2.20)
∂ψ1 ∂ψ1 ∂ψ1
dψ1 = dx + dy + dz = 0
∂x ∂y ∂z
but along any integral curve we have dξ~ ∼ E, thus we get
∂ψ1 ∂ψ1 ∂ψ1
P+ Q+ R=0,
∂x ∂y ∂z
but then, it is clear that any function of both characteristic curves is also
a solution
Φ(ψ1 , ψ2 ) = 0 , (2.21)
and it is equivalent to say that the two integration constants that appear in
Eq. (2.20) are not independent (this fact used in example 1 of this chapter
above).
Example 2.2 Find the two characteristic curves of the equation

∂z ∂z
z + = 1.
∂x ∂y
The equation of the characteristic curves are
dx
= dy = dz
z
which have the solutions

ψ1 = z − y = c1
and
ψ1 = z 2 − 2x = c2 .
Thus, the general integral surface of the equation can be expressed as
Φ(c1 , c2 ) = 0
or
z 2 − 2x = c2 (c1 ) ,
where c2 is arbitrary function of c1 (or other way around).
The solution of the Cauchy initial problem will be given in the next theo-
rem.
Theorem 2.1. Consider the first order quasi-linear partial differential

equation (2.15), where P, Q, R have continuous partial derivatives with re-
spect x, y, z. Suppose that along the initial curve x = x0 (s), y = y0 (s), the
initial values z = z0 (s) are prescribed, x0 , y0 , z0 being continuously differ-
entiable function for 0 ≤ s ≤ 1. Furthermore, let these function satisfy the
expression
dy0 dx0
P (x0 (s), y0 (s), z0 (s)) − P (x0 (s), y0 (s), z0 (s)) 6= 0. (2.22)
ds ds
Then, there exist one and only one solution z(x, y) defined in some
neighborhood of the initial curve, which satisfies the partial differential equa-
tion and the initial condition
z(x0 (s), y0 (s)) = z0 (s).
Proof. From the existence and uniqueness theorem for ordinary differ-
ential equations we may solve equations (2.16b) for a unique family of
characteristic
x = x(x0 , y0 , z0 , t) = x(s, t), (2.23a)
y = y(x0 , y0 , z0 , t) = y(s, t), (2.23b)
z = z(x0 , y0 , z0 , t) = z(s, t) (2.23c)
whose derivatives with respect to the parameter s, t are continuous such
that they satisfy the initial conditions
x(s, 0) = x0 (s), y(s, 0) = y0 (s), z(s, 0) = z0 (s).
We note that the Jacobian

∂(x, y) xs xt dx0 dy0
= det = Q− P 6= 0
∂(s, t) t=0
ys yt t=0 ds ds
because of the condition (2.22). Thus, in Eq. (2.23) we many solve for s, t
in terms of x, y in the neighborhood of the initial curve t = 0, obtaining
from Eq. (2.23) a candidate for solution
φ(x, y) = z(s(x, y), t(x, y)). (2.24)
This function satisfies the following initial conditions
φ(x, y)t=0 = z(s, 0) = z0 (s)
and satisfies also the differential equation (2.15) since one has

∂z ∂z ∂z ∂z
P φx + Qφy = P sx + tx +Q sy + ty
∂s ∂t ∂s ∂t
∂z ∂z
= (P sx + Qsy ) + (P tx + Qty )
∂s ∂t
∂z dx dy ∂z dx dy
= sx + sy + tx + ty
∂s dt dt ∂t dt dt

∂z ds ∂z dt
= +
∂s dt ∂t dt
∂z ∂z
= (0) +
∂s ∂t
∂z
=
∂t
= R(x, y, z).
Moreover, φ(x, y) is unique. Suppose that φ0 (x, y) is any other solution
satisfying the initial condition and x0 , y 0 is an arbitrary point in the neigh-
borhood of the initial curve. We consider the characteristic curve on the
surface φ
x = x(s0 , t0 ), y = y(s0 , t0 ), z = z(s0 , t0 )
where s0 = s0 (x0 , y 0 ). At t = 0, this curve passes through both surfaces
since it passes through the initial curve at the point
x(s0 , 0) = x0 (s), y(s0 , 0) = y0 (s), z(s0 , 0) = z0 (s).
But if a characteristic curve has one point in common with an integral
surface, it lies entirely on the surface. Thus, the characteristic curve lies on
both surfaces, and in particular for t0 , we have finally
φ0 (x0 , y 0 ) = φ0 (x0 (s0 , t0 ), y 0 (s0 , t0 )) = z(s0 , t0 ) = φ(s0 , t0 ).
When the initial curve is given by the intersection of two surfaces

Φ1 (x, y, z) = 0, Φ(x, y, z) = 0, (2.25)
we can obtained the solution using Eq. (2.20) and Eq. (2.25) to determine
c1 , c2 and the functionality Φ of Eq. (2.21). Note that the condition Eq.
(2.22) implies the condition that the initial conditions must not be given
on any characteristic curve.
Example 2.3 Find the integral surface of the equation

∂z ∂z
x −y =0 (2.26)
∂y ∂x
passing through the curve
x = 0, y = z2. (2.27)
We have three ways to determine the Cauchy problem, one is using the
parametric solutions, the other is using relation (2.21), and the last one is
using Eq. (2.25) (which is equivalent to the latter one). Let us explore each
of these ways.
i) Parametric solutions. The characteristic curves are given by equations
dx dy dz
= −y, = x, =0 (2.28)
dt dt dt
whose solutions are
x = aeit + be−it , y = i(aeit − be−it ), z = c1 , (2.29)
where a, b, c are constants. The initial curve is given as
x0 (s) = 0, y0 (s) = s, z0 (s) = s2 . (2.30)
Using these in Eq. (2.29), we obtain the family of characteristic
x(s, t) = s sin t, y(s, t) = s cos t, z(s, t) = s2 . (2.31)
The Jacobian is
∂(x, y)
=s
∂(s, t) t
then for s 6= 0, we have
s2 = x2 + y 2 ,
and the integral surface, from Eq. (2.31), is given by
z(x, y) = x2 + y 2 .
2.4. Quasi-linear PDEFO for Functions Defined in Ω ⊂ Rn 45
ii) Using the functionality of one characteristic with respect the other.
The equations for the characteristic curves are
dx dy dz
= =
−y x 0
whose solutions are
x 2 + y 2 = c1 , z = c2 . (2.32)
According with Eq.(2.21)
z = c2 (c1 ) = c2 (x2 + y 2 ),
and using the initial conditions (2.27), we obtain
s2 = c2 (s2 ) or c2 (x2 + y 2 ) = x2 + y 2
then, the solution is
z(x, y) = x2 + y 2 .
iii) Using the system (2.25) and (2.20).
We have the following set the equations
x = 0, z = y2 , x2 + y 2 = c1 and z = c2
where we obtain
c1 = y 2 = z = c2
then, the solution is again the same
z(x, y) = x2 + y 2 .
2.4 Quasi-linear PDEFO for Functions Defined in Ω ⊂ Rn
A quasi-linear PDEFO defined in Ω ⊂ Rn may be written as

n
X ∂z
Ri (x, z) = Q(x, z), (2.33)
i=1
∂xi
where x = (x1 , ..., xn ) is a point in Ω ⊂ Rn , z = z(x) is a function defined in

Ω with scalar values (in R) which is at least once continuously differentiable.
A solution z = f (x) defines a n-dimensional integral surface in Rn+1 whose
normal direction,
(∇z, −1)
√ ,
1 + ∇z · ∇z
(∇z = (z1 , ..., zn ) with zi = ∂z/∂xi ) is orthogonal to the vector field E

defined by
E = (R1 , ..., Rn , Q).
Thus, this field is tangent to the integral surface, and we may look for
family of curves on Rn+1 whose tangent direction, dξ~ = (dx1 , ..., dxn , dz)
coincides with the direction of E.
These curves are called characteristic curves and are given by equations
dx1 dxn dz
= ... = = (2.34a)
R1 Rn Q
or the parametric equations
dxi dz
= Ri (x, z) for i = 1, ..., n = Q(x, z), (2.34b)
dt dt
where 00 t“ is the parameter which characterizes these curves. If initial con-
ditions at t = 0 are characterized by the parameters s∗ = (s1 , ..., sn−1 ) ∈
Rn−1 ,
xi0 (s∗ ) = xi (0) for i = 1, ..., n z0 (s∗ ) = z(0), (2.35)
the solution of (2.34b) would be expressed as
xi = xi (s∗ , t) for i = 1, ..., n z = z(s∗ , t). (2.36)
Thus, choosing the condition that the Jacobian of the transformation
between x and (s∗ , t) to be different from zero,

∂(x1 , ..., xn−1 , xn )
6= 0 (2.37a)
∂(s1 , ..., sn−1 , t) t=0
(equivalent to the condition (2.22)), the inverse transformation,
s∗ = s∗ (x), t = t(x), (2.37b)
is gotten, and the solution of (2.33) with the conditions (2.35) is obtained
as
ψ(x) = z (s∗ (x), t(x)) . (2.38)
The theorem of existence and uniqueness (theorem 2.1) is extended im-
mediately to this case.
On the other hand, from (2.34a) one may obtain n-characteristics

φi (x) = Ci for i = 1, ..., n
2.4. Quasi-linear PDEFO for Functions Defined in Ω ⊂ Rn 47
and the general solution of (2.33) can be written as

Φ (φ1 (x), ..., φn (x)) = 0, (2.39)
where the function Φ is arbitrary. Let us extend the theorem 2.1 to <n .
Theorem 2.2. Consider the first order quasi-linear partial differential

equation (2.33), where Ri , Q, have continuous partial derivatives with re-
spect xi . Suppose that along the initial hypersurface parametrized by
s∗ ∈ <n−1 , x0i = x0i (s∗ ) for i = 1, . . . , n − 1, and values z = z0 (s∗ )
are prescribed, x0i , z0 being continuously differentiable function on each
sj , j = 1, . . . , n − 1. Furthermore, let these function satisfy the expres-
sion Eq. (2.37a). Then, there exist one and only one solution z(x) defined
in some neighborhood of the initial hypersurface Γo , which satisfies the par-
tial differential equation and the initial condition z(x0 (s∗ )) = z0 (s∗ ).
Proof. From the existence and uniqueness theorem for ordinary differ-
ential equations we may solve equations Eq. (2.34b) given the initial condi-
tions Eq. (2.35). Their solutions Eq. (2.36) have derivatives with respect
to the parameter sj for j = 1, . . . , n − 1 and t which are continuous such
that they satisfy the initial conditions
xi (s∗ , 0) = x0i (s∗ ) , for i = 1, . . . , n and z(s∗ , 0) = z0 (s∗ ) .
Since one has the condition Eq. (2.37a) is satisfied, one can have the inverse
relation Eq. (2.37b). Therefore, the function Eq. (2.38) is our candidate
for solution of the problem. Indeed, one has (using Eq. (2.34a) and the
linear independence of the variables sj and t)
n
"n−1 #
X ∂ψ X X ∂z ∂sl ∂z ∂t
Rk = Rk +
∂xk ∂sl ∂xk ∂t ∂xk
k k=1 l=1
n n−1 n
X X ∂z ∂sl ∂z X ∂t
= Rk + Rk
∂sl ∂xk ∂t ∂xk
k=1 l=1 k=1
n n−1 n
X X ∂z ∂xk ∂sl ∂z X ∂xk ∂t
= +
∂sl ∂t ∂xk ∂t ∂t ∂xk
k=1 l=1 k=1
n−1 n
!
X ∂z X ∂xk ∂sl ∂z dt
= +
∂sl ∂t ∂xk ∂t dt
l=1 k=1
n−1
X ∂z dsl ∂z
= +
∂sl dt ∂t
l=1
=Q
Therefore, the function Eq. (2.38) is a solution of the equation Eq. (2.33).
Now, this solution is unique since assume there is another one ψ 0 (x) sat-
isfying Eq. (2.33) which pass through the hypersurface Γo (s∗ ). Then,
since one must have that ψ 0 (x0 )|t=0 = ψ(x)|t=0 for any x0 , x ∈ <n and the
solutions are defined along the whole characteristics, one must have that
ψ 0 (x) = ψ(x) for any x ∈ <n .
Example 2.4 Find the integral surface of the equation

n
X ∂z
xi = αz, (2.40)
i=1
∂xi
where α ∈ R is given. The equations for the characteristic are
dx1 dxn dz
= ... = = . (2.41)
x1 xn αz
Taking the nth-term together with each previous one (this selection is
arbitrary), one gets the n − 1 characteristics
xi
Ci = for i = 1, ..., n − 1.
xn
Taking the last two terms of (2.41), one gets the characteristic
z
Cn = α .
xn
Thus, the general solution of (2.40) is given by

x1 x2 xn−1 z
Φ , , ..., , = 0.
xn xn xn xn
This solution can also be written as

x1 x2 xn−1
z = xαn Ψ , , ..., (2.42)
xn xn xn
This function is an homogeneous function of order α since for any λ ∈ R,
it follows that
z(λx) = λα z(x).
2.5. Problems 49
2.5 Problems
2.1 Make the classification of the following partial differential equations

2 2
1/2 ∂ z
i) (x1 − x2 ) = 0, z = z(x1 , x2 )
∂x2i
∂z ∂z
ii) x1 ez + x2 + x21 x2 z = 0 z = z(x1 , x2 )
∂x2 ∂x1
∂2z ∂z ∂z
iii) (x21 + x22 + x23 ) 2 + + b(x1 ) = 0, z = z(x1 , x2 , x3 )
∂x1 ∂x2 ∂x3
∂u ∂u
iv) +u = 0 u = u(x, t)
∂t ∂x
∂z ∂z
v) x1 − x2 = mz z = z(x1 , x2 ) .
∂x2 ∂x1
2.2 Demonstrate that the following PDEFO are linear
∂L ∂L
i) y −y = 0 , L = L(x, y)
∂x ∂y
∂u ∂u ∂u
ii) x2 + y2 + z2 = πz , u = u(x, y, z)
∂x ∂y ∂z
∂K ∂K ∂K
iii) v + F (x, v, t) + = 0 , K = K(x, v, t)
∂x ∂v ∂t
2.3 Prove that the function φ is solution of its respective PDEFO
1 ∂z ∂z
i) φ(x, y) = (x2 + y 2 ) ; x −y =0
2 ∂y ∂x
1 ∂z ∂z
ii) φ(x, y) = (x2 − y 2 ) , x +y =0
2 ∂y ∂x
vy vx
iii) φ(x, y, vx , vy ) = A1 x, y, , vx vx + A2 x, y, vy , vy
vx vy
∂L ∂L
vx + vy + L = 0 , L = L(x, y, vx , vy )
∂vx ∂vy
2.4 Find the integral surface of the following PDEFO:
∂z ∂z
i) x +y = αz , z = z(x, y)
∂x ∂y
∂z ∂z
ii) y 2 − x3 = 0 , z = z(x, y)
∂x ∂y
∂u ∂u
iii) + = 0 , u = u(x, t)
∂t ∂x
∂z ∂z
iv) y α + xα+1 = αz , z = z(x, y), α ≥ 1 .
∂x ∂y
2.4 Find the integral surfaces of the following PDEFO which pass through
the respective curve Γo , and determine whether or not Γo is a characteristic
curve of the equation:
∂z ∂z
i) x +y = αz , Γo : {y = x, z = x/2}
∂x ∂y
∂L
ii) − L = K(x, v), Γo : {x = v, L = 0}
∂v
∂z ∂z
iii) x −y = αz , Γo {x2 + y 2 = 4, z = x2 }
∂y ∂x
1 ∂u ∂u
iv) + = f (x)u , Γo : {t = 0, u = 0}
c ∂t ∂x
∂z ∂z p
v) x1/3 − =0, Γo (s) : {xo (s) = 1, yo (s) = s, zo (s) = 1 − s2 } .
∂y ∂x
2.5 Giving the curve of initial data, Γo , show that the following problems
have not solutions:
∂z 1 ∂z
i) − = 0 , Γo : {x + y 2 = 1, z = x2 }
∂x y ∂y
∂L
ii) v − L = K(x, v) , Γo : {x = 0, L = v 2 /2}
∂v
∂z ∂z
iii) x2 − y2 = x2 z , Γo : {y = (1 − x3 )1/3 , z = x2 e−x }
∂y ∂x
∂K ∂K
iv) v −x = 0 , Γo : {x2 + v 2 = 1/3, K = v 2 /2} .
∂x ∂v
2.6 Demonstrate that the solution of the PDEFO defined in <n by
n
X ∂z
xi = −αz , α≥0
i=1
∂xi
such that pass through the hyperplane Γo : {x1 = 1}, having the value there
as z = φ(x2 , . . . , xn ), with φ being any differentiable function, is given by
z(x) = φ(x1 , . . . , xn )eα(1−x1 ) .
2.7 Demonstrate that the solution of the PDEFO z 2 (∂z/∂x)−∂z/∂y = z,

such that at y = 0, z(x, 0) = φ(x) is an invertible function in the domain
Ω ⊂ <2 defined by the equation, is solution of the following transcendental
algebraic equation
z 2 (1 − e2y ) + 2φ−1 (zey ) − 2x = 0

2.5. Problems 51
2.8 Demonstrate that the solution of the PDEFO ut − uux = 0 defined

in <2 , u = u(x, t), such that u(x, 0) = φ(x), has the implicit solution

u(x, t) = φ x − u(x, t)t .
In addition, suppose that φ(x) is of the form
1 2 2
φ(x) = √ e−x /2σ .
σ 2π
Then, make a plot of the solution (σ = 1).
2.9 Find the solution of the PDEFO defined in <3 by

∂u ∂u ∂u
x −y +u =0,
∂y ∂y ∂z
with u = u(x, y, z) and such that u(0, y, z) = ψ(y, z), and show that this
solution is given implicitly as
!
p
2 2
x
u(x, y, z) = φ x + y , ±u(x, y, z) arcsin p +z
x2 + y 2
2.10 Find the solution in parametric form of the PDEFO defined in <3
by
∂z ∂z ∂z
x1 + x2 + x3 = z2
∂x1 ∂x2 ∂x3
which pass through the initial surface defined as Γo (s∗ ) : {x1o (s∗ ) =
1, x2o (s∗ ) = s1 , x3o (s∗ ) = s2 , zo (s∗ ) = s1 + s2 }, and show that the solu-
tion can be written as
x2 + x3
z(x) =
x1 − (x2 + x3 ) ln x1
2.11 Find the solution of the quasi-linear PDFEO
∂u ∂u ∂u
u +y +z = u2
∂x ∂y ∂z
such that it passes through the initial surface Γo : {x = 0, u = y + z}.
2.6 References
2.1 F. John, Partial Differential Equations, Springer-Verlag, 1978.

Chap.1.
2.2 L. Elsgoltz, Ecuaciones Diferenciales y Cálculo Variacional, Edi-

ciones Cultura Popular, 1975. Pages 248-260.
2.3 R. Courant and D. Hilbert, Methods of Mathematical Physics, John

Wiley & Sons, 1962. Vol. II chapters I and II.
Chapter 3
Physical Applications I
In this chapter we illustrate the uses of the linear partial differential equa-
tions of first order in several topics of Physics. The purpose of this chapter
is to motivate the importance of this branch of mathematics into the phys-
ical sciences. In most of the applications, it is not intended to fully develop
the consequences and the theory involved in the applications, but usually
we point out some reference where the physical theory can be seen.
3.1 Mechanics
As we saw in chapter 1.6, there is a connection between Newton’s equations

of motion (1.58), Euler’s equation (1.62), the Lagrangian (1.61) and the
Hamiltonian (1.64) of the system. One of the most important objectives
of Mechanics is not only to give the solution of the equation of motion in
either form (1.58), (1.65) or (1.76), but also to find the constant of motion
of the system. In terms of Eq. (1.58) and for time independent forces
(autonomous systems), we have
dvi
= fi (q, v), i = 1, ..., N (3.1a)
dt
and
dqi
= vi , i = 1, ..., N (3.1b)
dt
By a constant of motion we mean any surface K = K(q, v) defined in
2N - dimensional space (q, v) such that is normal direction projection on
this space is the perpendicular to the vector field (1.59b),
E = (v, f ), (3.2)
53
54 Physical Applications I
where v = (v1 , ..., vN ) and f = (f1 , ..., fN ). That is, the function K satisfies
the linear partial differential equation
N
X ∂K ∂K
vi + fi (q, v) = 0, (3.3)
i=1
∂qi ∂vi
and the total time derivative operator along E is defined as
N
d X ∂ ∂
= vi + fi (q, v) . (3.4)
dt i=1
∂qi ∂vi
As we saw in chapter 2, the solution of Eq. (3.3) is given by
K = K(c1 , ..., c2N −1 ), (3.5)
where ci (i = 1, ..., 2N − 1) are the characteristic curves, which are solutions
of equations
dq1 dqN dv1 dvN
= ... = = = ... = . (3.6)
v1 vN f1 fN
Example 3.1. Find the constant of motion of one dimensional single par-
ticle of mass ”m” moving in a dissipative medium whose frictional force is
proportional to the velocity squared.
The equation of motion can be written as the following dynamical sys-
tem, Eq. (1.84),
dv α
= − v2
dt m
and
dq
= v,
dt
where α is the friction constant and m is the mass os the particle, and
v > 0. The equation of the characteristic curve is given by
dq dv
=
v −(α/m)v 2
whose solution gives us the characteristic curves
c = veα/mq . (3.7)
Thus, the constant of motion is given by an arbitrary function of “c”,
K = K(veα/mq ) • (3.8)
As we can see from Eq. (3.5) or Eq. (3.8), there exists a non-numerable
3.1. Mechanics 55
set of constants of motion for a given physical system. One possible way
to select a proper function is using the criteria of reducibility. It consists
in choosing the functionality K in such a way that we can get the known
form of the energy (E = mv 2 /2 + V (q)) or some other constant of motion
when the parameters, which characterize the nonconservative force, go to
zero. In the above example if we take
1 2 1
mc = mv 2 e2αq/m ,
K= (3.9)
2 2
we obtain the usual energy of a free particle for α going to zero. Thus, one
can use this constant of motion as the constant of motion which contains
the proper dissipative dynamics of the system.
If the Hamiltonian in the expression (1.58) is substituted by the constant

of motion (3.5) and relation (1.63) is used. Then, given the constant of
motion, we obtain the following linear partial differential equation for the
Lagrangian
N
X ∂L
K(q, v) = q̇ − L. (3.10)
i=1
∂ q̇
The equations for the characteristic are
dq1 dqN dq̇1 dq̇N dL
= ... = = = ... = = . (3.11)
0 0 q̇1 q̇N L+K
From the first 2N terms, we obtain
Ci = qi i = 1, ..., N (3.12a)
and
Cij = q̇i /q̇j (3.12b)
such that
Cji Cij = 1 i 6= j = 1, ..., N. (3.13)
Now, making use of Eq. (3.12) and Eq. (3.13)in the last term of Eq.
(3.11) and integrating with any of the N -terms of its left hand side, we get
N
X
L= Ai (C, C(i) )q̇i +
i=1
N Z q̇i
1 X K(C, C1i , C2i , ..., ξ, Ci+1,i , ..., CiN )
+ q̇i dξ (3.14)
N i=1 ξ2
where Ai are arbitrary functions of the characteristics C = (C1 , ..., CN ) and

C(i) = (C1i , C2i , ..., Ci−1,i , Ci+1,i , ..., CN i ). The first on the right hand side
of Eq. (3.14) is the solution of the homogeneous part of Eq. (3.10), that is
when K is equal to zero. This can be seen using the fact that
N
X ∂Ai
q̇i =0 (3.15)
i=1
∂ q̇i
An the second term on the right hand side of Eq. (3.14) represents the
solution of the inhomogeneous equation (3.10). The factor N appearing in
this term is necessary to accomplish this. The generalized linear momentum
(1.63) is
1 q̇j K(C, C1j , ..., Cj−1,j , ξ, Cj+1,j , ..., CjN )
Z
(i) K
pj = Aj (C, C ) + dξ +
N ξ2 N q̇j
and after making an integration by parts, we get
1 q̇j ∂K(C, C1j , ..., Cj−1,j , ξ, Cj+1,j , ..., CjN ) dξ
Z
pj = Aj + (3.16)
N ∂ξ ξ2
Example 3.2. Find the Lagrangian and the generalized linear momentum
for the system described in the Exam. (3.1) with the constant of motion
given by Eq. (3.9).
According to Eq. (3.14), the Lagrangian for one-dimensional au-

tonomous systems can be calculated through
Z v
K(q, ξ)
L(q, v) = A(q)v + v dξ, (3.17a)
ξ2
if
1
K(q, v) =mv 2 e2αq/m (3.17b)
2
is the constant of motion, we have the Lagrangian and the general momen-
tum as
1
L = A(q, v)v + mv 2 e2αq/m (3.18)
2
and
p = A(q, v) + mve2αq/m (3.19)
The first term of the right hand side of Eq. (3.17a) is the total time deriva-
tive of a function depending on the variable q. As it was pointed out for
Eq. (1.72), this term gives us an equivalent Lagrangian therefore, it can be
3.1. Mechanics 57
neglected •
Let us now apply the operator (3.4) to expression (3.10) to look for a
relation between expression (3.10) and Euler’s equation (1.62),
N X N N
∂2L ∂2L
X
dK X ∂L
= q̇i q̇n + fn q̇i − q̇i ,
dt i=1 n=1
∂ q̇i ∂qn ∂ q̇ i ∂ q̇ n i=1
∂ q̇i
then after rearranging terms

N N !
dK X X ∂ ∂L ∂ ∂L ∂L
= q̇i q̇n + fn −
dt i=1 n=1
∂qn ∂ q̇i ∂ q̇n ∂ q̇i ∂ q̇i
and according to definition (3.14), the above expression can be written as

N
dK X d ∂L ∂L
= q̇i − . (3.20)
dt i=1
dt ∂ q̇i ∂qi
Let V and E be the vectors defined as
V = (q̇1 , ..., q̇n )
and

d ∂L ∂L d ∂L ∂L
E= − , ..., − ,
dt ∂ q̇1 ∂q1 dt ∂ q̇n ∂qn
then, Eq. (3.20) expresses the fact that whenever K is a constant of motion
of the system, the orthogonality relation
V·E=0 (3.21)
is always satisfied. We must notice that this does not guarantee the ex-
istence of the Lagrangian except for the one-dimensional case. For q̇ 6= 0,
K in a one-dimensional case, K is a constant of the motion if and only if
L satisfies the Euler’s equation (1.62). This means that we can always find
the Lagrangian by using Eq. (3.17). For the n-dimensional case, we can use
expression (3.14) and Euler’s equation to find restrictions on the functions
Ai and the constant of motion K in order for the L function to represent
the Lagrangian of the system (3.1). Thus, developing the term
N
X ∂ ∂ ∂L ∂L
q̇n + fn − =0
n=1
∂qn ∂ q̇n ∂ q̇l ∂ql
with L given by the expression (3.14), we have

N X N N Z q̇l
X ∂Al ∂Al ∂Ai 1 X ∂K dξ
q̇n + fn − q̇i + q̇n 2
n=1
∂qn ∂ q̇n i=1
∂ql N n=1
∂qn ξ
N Z q̇i N
1 X ∂K dξ 1 X ∂Al ∂Al
− q̇i + q̇n + f n = 0,
N i=1 ∂ql ξ 2 N q̇l n=1 ∂qn ∂ q̇n
and using the fact that K is a constant of motion, being Ai an arbitrary
independent functions of K, we get
N
"Z #
q̇l Z q̇n
X ∂K dξ ∂K dξ
q̇n − =0 (3.22a)
n=1
∂qn ξ 2 ∂ql ξ 2
and
N X N
X ∂Al ∂An ∂Al
q̇n + + fn =0 (3.22b)
n=1
∂qn ∂ q̇l n=1
∂ q̇n
for every l = 1, ..., N . So, if the conditions (3.22) are satisfied, the function
(3.14) represents a Lagrangian for the system (3.1).
Example 3.3. Find a constant of the motion for a particle moving in three
dimensional space under a field of forces derivable from a potential function
φ = φ(q). Verify the relation (3.22) and find a Lagrangian.
The equations of motion (3.1) are

dvi 1 ∂φ
=− i = 1, 2, 3 (3.23a)
dt m ∂qi
and
dqi
= vi i = 1, 2, 3 (3.23b)
dt
where m is the mass of the particle. The equations for the characteristic
curves of the constant of motion (3.6) are
dq1 dq2 dq3 mdv1 mdv2 mdv3
= = = = =
v1 v2 v3 −(∂φ/∂q1 ) −(∂φ/∂q2 ) −(∂φ/∂q3 )
which can be completed to yield the exact differential

1
d φ + m(v12 + v22 + v32 ) = 0.
2
Then, a possible constant of motion is
1
K = m(v12 + v22 + v32 ) + φ. (3.24)
2
3.1. Mechanics 59
FromEq. (3.23a), wesee that

∂φ d 1 d 1
3vl = −3 mvl2 = − m(v12 + v22 + v32 ) , (3.25)
∂ql dt 2 dt 2
but taking the time derivative of (3.24), we obtain
3
d 1 dφ X ∂φ
m(v12 + v22 + v32 ) = − =− q̇n
dt 2 dt n=1
∂ql
so we obtain for Eq. (3.25)
X 3
∂φ ∂φ
3q̇l = q̇n for l = 1, 2, 3. (3.26)
∂ql n=1
∂qn
Using Eq. (3.25) in Eq. (3.22a), we have
3 " 3 #
X ∂φ 1 ∂φ 1 1 X ∂φ
∂φ
q̇n − + = − q̇n + 3q̇l
n=1
∂qn q̇l ∂ql q̇n q̇l n=1
∂qn ∂ q̇l
which according to Eq. (3.26) is zero. Choosing the functions Ai = 0 (i =
1, 2, 3), the condition Eq. (3.22b) is also satisfied. Using now Eq. (3.24) in
the expression (3.14), we get
1
L = m(v12 + v22 + v32 ) − φ • (3.27)
2
Now, using the calculated generalized linear momentum (3.16), if we
can make
q̇i = q̇i (q, p) i = 1, ..., N (3.28)
then, we could use these functions in Eq. (3.5) to obtain the Hamiltonian
associated with the system (3.1) as
H(q, p) = K(c1 (q, q̇(q, p)), ..., c2n−1 (q, q̇(q, p))). (3.29)
Example 3.4. Calculate the Hamiltonian for the Example (3.2) and Ex-
ample (3.2).
In the Example (3.2), we take A(q) = 0 in expression (3.19) to obtain
p = mq̇e2αq/m .
Then, the inverse relation is
p
q̇ = e−2αq/m ,
m
and substituting this expression in Eq. (3.17b), the Hamiltonian becomes
p2 −2αq/m
H= e .
2m
In Example (3.3), we get from Eq. (3.27) and Eq. (1.63)
pi = mq̇i i = 1, 2, 3.
Substituting this in Eq. (3.24), we obtain the Hamiltonian
1 2
H= (p + p22 + p23 ) + φ.
2m 1
3.2 Angular Momentum in Quantum Mechanics
In classical mechanics, the angular momentum vector of a particle with

respect to a reference system S is defined as
L=r×p (3.30)
where r is the position vector of the particle with respect to S and p is

its linear momentum. In quantum mechanics, the components of the linear
momentum are represented by the linear operators
∂ ∂ ∂
px = −i~ , py = −i~ , and pz = −i~ , (3.31)
∂x ∂y ∂z
where ~ is Planck’s constant divided by 2π. Hence, the operator associated
to the components of the angular momentum vector are given in cartesian
coordinates (x, y, z) as

∂ ∂
Lx = −i~ y −z , (3.32a)
∂z ∂y

∂ ∂
Ly = −i~ z −x , (3.32b)
∂x ∂z

∂ ∂
Lz = −i~ x −y , (3.32c)
∂y ∂x
and in spherical coordinates (r, θ, φ) they are given as follows

∂ ∂
Lx = i~ sin φ + cot θ cos φ , (3.33a)
∂θ ∂φ

∂ ∂
Ly = i~ − cos φ + cot θ sin φ , (3.33b)
∂θ ∂φ
∂
Lz = −i~ , (3.33c)
∂φ
where the relation between cartesian and spherical coordinates is as follow-
ing
x = r sin θ cos φ (3.34a)

y = r sin θ sin φ (3.34b)
y = r cos θ (3.34c)
The square of the total angular momentum is defines as
L2 = L2x + L2y + L2z (3.35)

3.2. Angular Momentum in Quantum Mechanics 61
and is expressed in terms of spherical coordinates as

1 ∂2

1 ∂ ∂
L2 = −~2 sin θ + . (3.36)
sin θ ∂θ ∂θ sin2 θ ∂θ2
The operators (3.32) or (3.33) satisfy the following commutation rules
[Li , Li ] = 0 i = x, y, z, (3.37a)
and
[Lx , Ly ] = i~Lz , [Ly , Lz ] = i~Lx , [Lz , Lx ] = i~Ly (3.37b)
where the commutation between two operators A and B is a bilinear oper-
ation defined as
[A, B] = AB − BA (3.38)
and it has the property
[AB, C] = A[B, C] + [A, C]B (3.39)
2
for every A, B, C operators. Let us calculate [L , Lx ]. Using the relations
(3.35) and (3.37b)
[L2 , Lx ] = [L2x , Lx ] + [L2y , Lx ] + [L2z , Lx ]
= −i~(Ly Lz + Lz Ly ) + i~(Lz Ly + Ly Lz ) = 0
and in the same way
[L2 , Ly ] = [L2 , Lz ] = 0. (3.40)
The relation(3.37) tell us that the set Lx , Ly , Lz from the Lie algebra
associated to the group of rotations in three dimensional space, called O(3),
or also the Lie algebra of 2 × 2 complex matrix with determinant equal to
one (so called SU (2)). This group leaves invariant the usual Euclidean
metric in this space (ds2 = dx2 + dy 2 + dz 2 ). Defining the new operators
L+ = Lx + iLy (3.41a)
and
L− = Lx − iLy , (3.41b)
the relations (3.37) and (3.40) take the standard form
[Lz , L+ ] = ~L+ , [Lz , L− ] = −~L− , [L+ , L− ] = 2~Lz (3.42)
and
[L2 , L+ ] = [L2 , L− ] = [L2 , Lz ] = 0. (3.43)
In spherical coordinates and using Eq. (3.33), L+ and L− are written

as

iφ ∂ ∂
L+ = ~e + i cot θ (3.44)
∂θ ∂φ
and

∂ ∂
L− = ~e−iφ − + i cot θ . (3.45)
∂θ ∂φ
Let us choose the projection Lz of the angular momentum vector to do
the quantization. Because of relation(3.43), we know that the operator L2
and Li have common eigenfunctions, but different eigenvalues, that is
L2 Uλm = ~2 λUλm (3.46)
and
Lz Uλm = ~mUλm , (3.47)
with the normalization condition
Z
†
Uλm Uλm dΩ = 1, (3.48)
where dΩ = sin θdθdφ is the differential of the solid angle, Uλ† represents
the transpose conjugated operation with the properties: i) f † = f ∗ , ii)
(A + αB)† = A† + α∗ B † , and iii) (AB)† = B † A† for any linear operators A
and B, any complex number α, and ny complex function f . This expression
is usually written as hλm|λmi = 1. Using Eq. (3.33c), the solutions of
Eq. (3.47) in spherical coordinates is
Uλm = fλ (θ)eimφ ,
and in cartesian coordinates, using Eq. (3.32c), we have to solve the follow-
ing linear partial differential equation
∂Uλm ∂Uλm
x −y = imUλm
∂y ∂x
from which the equations for the characteristic curves are
dy dx dUλm
= = .
x −y imUλm
From the first two terms, we obtain the characteristic
c = x2 + y 2 ,
and using this expression in the third term, the resulting equation is
Z Z
dy dUλm
im =
(c − y 2 )1/2 Uλm

y
Uλm = Aλ (x2 + y 2 ) exp im arcsin .
(x2 + y 2 )1/2
If m ∈ N ,the solution is periodic with respect the angle φ. To find
the complete solution and the eigenvalues λ of the operator L2 , let us use
the algebraic relations (3.42) The action of the operators L+ and L− on
the function Uλm can be know through the action Lz L± Uλm and using
Eq. (3.42) and Eq. (3.47). The operator (3.32) and (3.36) are Hermitian.
One say that an operator A is Hermitian if it satisfies the relation
Z Z
†
Uλm (AUλm )dΩ = (AUλm )† Uλm dΩ, (3.49a)
that is A† = A. Note, however, that the operators L+ and L− are not

Hermitian, in fact
L†+ = L− (3.49b)
and
L†− = L+ . (3.49c)
Adding and subtracting the operator L± Lz and using Eq. (3.42), if
follows
Lz L± Uλm = ([Lz , L± ] + L± Lz )Uλm = (m ± 1)~(L± Uλm ),
in other words, the function L± Uλm is also an eigenvector of the operator
Lz , but with eigenvalues m ± 1,
Lz (L± Uλm ) = ~(m ± 1)(L± Uλm ),
this means that the function L± Uλm is proportional to the function Uλm±1 .
if a±
λm is the constant of proportionality, thus we can write
L+ Uλm = a+
λm Uλm+1 (3.50)
and
L− Uλm = a−
λm Uλm−1 . (3.51)
Since the quantity
Z
†
Uλm (L2x + L2y )Uλm dΩ ≥ 0
is a non-negative real number and using Eq. (3.35), Eq. (3.46), Eq. (3.47)
and the normalization condition (3.48), we have
Z Z
† †
Uλm (L2x + L2y )Uλm dΩ = Uλm (L2 − L2z )Uλm dΩ = ~2 (λ − m2 ) ≥ 0.
Therefore, m2 ≤ λ, in other words the number the eigenvalues m in

finite. This means that there must be two extrema values m1 and m2 such
that, according to Eq. (3.50) and Eq. (3.51), we must have
L+ Uλm1 = 0 (3.52a)
and
L− Uλm2 = 0. (3.52b)
On the other hand, using Eq. (3.42), the action of L+ L− and L− L+ on
the function Uλm can be calculated as
L+ L− Uλm = ([L+ , L− ] + L− L+ )Uλm = (2~Lz + L2x + L2y − ~Lz )Uλm

= (~Lz + L2 − L2z )Uλm = ~2 (λ − m2 + m)Uλm (3.53a)
and similarly
L− L+ Uλm = ~2 (λ − m2 − m)Uλm . (3.53b)
Thus, for the eigenvalues m2 and m1 , if follows that
L+ L− Uλm2 = 0 (3.54a)
and
L− L+ Uλm1 = 0. (3.54b)
Due to Eq. (3.53), we get the algebraic equations
λ + m2 − m22 = 0 (3.54c)
and
λ − m1 − m21 = 0. (3.54d)
The solutions of these equations brings about two possibilities: m1 =
m2 − 1 or m1 = −m2 . The first is not possible since we have chosen m1 >
m2 . Therefore, the second solution, m1 = −m2 , is the only possibility.
Now, from Eq. (3.50) and Eq. (3.52), we know that the difference m1 − m2
is a integer number (we can go up by applying step by step the operator L+
to a given function Uλm ). For this reason 2m1 ∈ Z. Defining this maximum
value m1 as j, (m1 = j), the possible values of j are
j = 0, ±1/2, ±1, ±3/2, ±2, ... (3.55a)
and from Eq. (3.54d), the eigenvalues λ is given as
λ = j(j + 1). (3.55b)
On the other hand, since λ ≥ 0, one must have that j(j + 1) ≥ 0. So, one
finally gets its allowed values
j = 0, 1/2, ±1, ±3/2, ±2, ... (3.55c)
The function Uλm can be characterized as Ujm , where the maximum value
of m is m = j, and its minimum value is m = −j. Let us now calculate
−
the coefficients a+
jm and ajm appearing in Eq. (3.50) and Eq. (3.51). Using
Eq. (3.48) and Eq. (3.50), we have
Z Z
† 1
1 = Ujm+1 Ujm+1 dΩ = + 2 (L+ Ujm )† (L+ Ujm )dΩ
|ajm |
Z
1 †
= + 2 Ujm (L− L+ Ujm )dΩ.
|ajm |
Thus, the coefficient a+
jm can be calculated from the equation
Z
†
|a+
jm | 2
= Ujm (L− L+ Ujm )dΩ.
Now, using Eq. (3.53b) and Eq. (3.55b) above, it follows

|a+ 2 2 2
jm | = ~ (j(j + 1) − m(m + 1)) = ~ (j − m)(j + m + 1). (3.56a)
In a similar way, using the expression
Z
†
1 = Ujm−1 Ujm−1 dΩ,
we get
|a− 2 2 2
jm | = ~ (j(j + 1) − m(m − 1)) = ~ (j + m)(j − m + 1). (3.56b)
We usually consider the coefficients and a+
jm a−
as real numbers. The
jm
resulting equations that we obtain from Eq. (3.55), Eq. (3.56), Eq. (3.46)
and Eq. (3.47) are the following
p
L+ Ujm = ~ (j − m)(j + m + 1)Ujm+1 , (3.57a)
p
L− Ujm = ~ (j + m)(j − m + 1)Ujm−1 , (3.57b)
~ p p
Lx Ujm = (j − m)(j + m + 1)Ujm+1 + ~ (j + m)(j − m + 1)Ujm−1 ,
2
(3.58a)
~ p p
Ly Ujm = (j − m)(j + m + 1)Ujm+1 − (j + m)(j − m + 1)Ujm−1 ,
2i
(3.58b)
L2 Ujm = ~2 j(j + 1)Ujm , (3.59a)
Lz Ujm = ~mUjm , (3.59b)

having the following conditions
L+ Uj,j = 0 (3.60a)
and
L− Uj,−j = 0. (3.60b)
The allowed values for the quantum number j are given by 3.55c, and
the quantum number m must be such that
−j ≤ m ≤ j. (3.61)
Therefore, for each j value, there are 2j + 1 possible values for the quan-
tum number m (there are 2j + 1 possible projections of the angular mo-
mentum in the z-axis). Due to Eq. (3.57a) and Eq. (3.58b), we see that
it is enough to know a single function Ujm to find the set of functions
{Uj,−j , Uj−j+1 , ..., Uj,j } through the application of the operators L+ and
L− to this given function. Given the eigenvalues j, the easiest functions to
obtain are Uj,j or Uj,−j because they satisfy Eq. (3.60). Foe example, using
Eq. (3.60b) and the operator (3.45), Uj,−j is the solution of the following
linear partial differential equation
∂Uj,−j ∂Uj,−j
− + i cot θ = 0. (3.62)
∂θ ∂φ
The equation for the characteristics are
dθ dφ dUj,−j
= = .
−1 i cot θ 0
From the first two terms, we get
Z Z
cot θdθ = i dφ,
and making the integration this

log sin θ − iφ = constant.
Then, the characteristic curve can be written as
c = e−iφ sin θ.
3.3. Heat Propagation between Two Superconducting Cables 67
The solution of Eq. (3.62) is an arbitrary function, Gj , of the above

characteristic curve,
Uj,−j = Gj (eiφ sin θ). (3.63)
This solution in completaly general for any 2j ∈ Z. However, for j = l ∈ Z
one can choose a polynomial of order l as a particular solution (Gl (c) = cl ),
i.e.
Ul,−l = a(e−iφ sin θ)l = ae−ilφ sinl θ,
where “a” is a constant of proportionality determined by the normalization
condition (3.48),
Z Z 2π Z π
†
1 = Ul,−l Ul,−l dΩ = |a| 2
sin2l+1 θdθdφ
0 0
22l+1 (l!)2
= |a|2 2π .
(2l + 1)!
Hence, the value of the constant of proportionality is given by

(2l + 1)!
a = 2l+1 ,
2 2π(l!)2
and the function Ul,−l is
r
1 (2l + 1)! −ilφ l
Ul,−l = l e sin θ. (3.64)
2 l! 4π
Any other function Ulm can be calculated now by applying l + m times
the operator L+ to the function Ul,−l . Using Eq. (3.57a) each time, we
obtain the functions
s " #
(−1)l+m (2l + 1)!(l − m!) imφ m dl+m sin2l θ
Ulm = e sin θ (3.65)
2l l! 4π(l + m)! d(cos θ)l+m
which are called “Spherical Harmonic Functions”, and the usual notation
for these functions is Ylm (θ, φ) or Ylm (θ, φ).
3.3 Heat Propagation between Two Superconducting Ca-

bles
A superconductor cable (scc) is characterized by a bounded surface con-

taining the origin defined in the space formed by the current (I) carried
by the scc (Cooper pairs of electrons), the magnetic field (H) applied to
scc, and the temperature (T ) of the medium the scc is imbibed. Below this
surface the resistivity (ρ) of the scc is zero, and there is not dissipation of
energy (heat) generated by flow the current in the cable. Above this sur-
face, the scc behaves like a normal cable whit a very high resistivity value.
The surface is called the “critical surface of the scc”, and each point on this
surface (Tc , Ic , Hc ) is called a “critical point of the scc” (see Fig. 3.1).
Fig. 3.1 Critical surface and a critical point.
When a region of a scc becomes normal (i.e. this region is above the
critical surface) because of some external perturbation, this normal zone
propagates along the scc, and the heat generated propagates transversely
to other scc. This phenomenon is known as “quenching (or quench)” on
the scc. Normally, a scc is made up of many superconducting wires (for
example NbTi) contained in a non superconducting matrix (usually Cu or
Al). The function of the non superconducting matrix is to make the current
flow within it when a quench occurs (because of its much lower resistivity
than that of superconducting wires at low temperature) in the scc, avoiding
the evaporation of the superconducting wires. Once the quench starts in a
scc, the increasing of the temperature where the normal zone first appeared
(which maybe the hottest point) as a function of time can be described by
the equation
dT1
(δcp ) = ρj 2 , (3.66)
dt
where (δcp ) is the average heat capacity of the scc, ρ is the average resis-
tivity, and j is the density of the current flowing in the cable. In between
two scc there is an insulation. For this reason the transversal propagation
3.3. Heat Propagation between Two Superconducting Cables 69
of heat is much slower. An approximate description of this transversal heat

propagation can be considered as follows: Let two scc’s be separated by
a distance L and suppose we know the variation of the temperature as a
function of time (T1 (t)) of the first scc where the quench appeared. At any
point “x00 between these scc’s, the density flow of heat in a distance interval
[x, x + ∆x] is given by qVh , where q is the density of heat, and Vh = ∆x/∆t
is the heat velocity. This heat flow is proportional to the change of the tem-
perature (∆θ) in this interval , and we assume it is inversely proportional
to the distance to the scc (x), i.e. −qVh = k∆θ/x, where k = k(θ) is the
function of proportionality, called “thermal conductivity”. Therefore, one
gets the relation
k(θ) ∆θ
−∆q = . (3.67)
x Vh
On the other hand, the variation of the density of heat in this region is
given by
∆q = (δcp )I ∆θ, (3.68)
where (δcp )I (θ) is the average heat capacity of the materials in between the
two scc’s which depends of the temperature. From Eq. (3.68), we get
∆θ k ∆θ
(δcp )I + = 0. (3.69)
∆t x ∆x
Taking the limits as ∆x and ∆t tend to zero, we obtain a quasi-linear
partial differential equation for our heat propagation model
∂θ k(θ) ∂θ
(δcp )I (θ) + = 0. (3.70)
∂t x ∂x
The equation for the characteristic of this equation are
dt dx dθ
= = .
(δcp )I (θ) k(θ)/x 0
From the last term, we find that one the characteristic curves is
C1 = θ(x, t),
and the substituting this in the other terms, it follows that
dt dx
= .
(δcp )I (C1 ) k(C1 )/x
We can integrate this last equation obtaining a constant (C2 ), characteriz-
ing the second characteristic curve, which depends on the fist characteristic
(C1 ),
(δcp )I (C1 ) 2
t− x = C2 (C1 )
2k(C1 )
or
(δcp )I (θ) 2
t− x = C2 (θ). (3.71)
2k(θ)
C2 can be determined through the boundary conditions, and this one can
be determined in the following way: we know through Eq. (3.66) how the
temperature varies in the first scc, that is, at x = 0. Therefore, we may
impose the boundary condition
θ(0, t) = T1 (t). (3.72)
Using this boundary condition in Eq. (3.71), we have
t = C2 (T1 (t)),
for all time t. But this implies that C2 is inverse function of T1 , and then
Eq. (3.71) looks like
(δcp )I (θ) 2
t− x = T1−1 (θ).
2k(θ)
Applying T1 to this equation, it follows that

(δcp )I (θ) 2
T1 t − x = θ. (3.73)
2k(θ)
The solution of Eq. (3.70) is given in the implicit form Eq. (3.73). If we
know the time at which the first scc quenches , it is of practical interest to
estimate the time that the heat will take to arrive to the other scc, inducing,
consequently, a quench. Suppose that these two scc are separated by a
distance L, and the temperature in the second scc as a function of time is
T2 (t), then, according to Eq. (3.73), this temperature is given by

(δcp )I (T2 (t)) 2
T1 t − L = T2 (t). (3.74)
2k(T2 (t))
If τ1 is the time taken for the first scc to reach the critical temperature
Tc1 and quench (T1 (τ1 ) = Tc1 ), and if τ2 is the time taken for the second
scc to reach the critical temperature Tc2 and quench (T2 (τ2 ) = Tc2 ). Thus,
evaluating Eq. (3.74) at the time τ2 , we get

(δcp )I (Tc2 ) 2
T1 τ2 − L = Tc2 ,
2k(Tc2 )
where, in general, Tc1 6= Tc2 since both scc are placed in regions where the
magnetic field may by different. However, normally both scc’s are close
enough to consider that the magnetic fields ere approximately the same,
then, we can consider Tc1 = Tc2 . In this way the right side of the above
3.4. Classical Statistical Mechanics in Equilibrium 71
equation represents the critical temperature of the first scc which happens
at the time τ1 . Therefore, it follows that
(δcp )I (Tc2 ) 2
τ2 − L = τ1
2k(Tc2 )
or
(δcp )I (Tc2 ) 2
τ2 = τ1 + L . (3.75)
2k(Tc2 )
The time τ1 can be estimated, integrating Eq. (3.66), as
1 Tc1 (δcp )c (T )
Z
τ1 = 2 dT,
j T0 ρ(T )
where T0 is the batch temperature.
3.4 Classical Statistical Mechanics in Equilibrium
As we saw in Eq. (1.82), Liouville’s theorem gives us the basic equations

to deal with many body problems in non equilibrium and in equilibrium
situations. To describe ensembles of particles in equilibrium one has to
solve the equation
3N
X ∂ρ ∂H ∂ρ ∂H
{ρ, H} = − = 0. (3.76)
i=1
∂qi ∂pi ∂pi ∂qi
The equations for its characteristic are
dq1 dq3N dp1 dp3N dρ
= ... = = = ... = = . (3.77)
∂H ∂H ∂H ∂H 0
− −
∂p1 ∂p3N ∂q1 ∂q3N
These equations can be arranged to have

∂H ∂H ∂H ∂H
dq1 + ... + dq3N + dp1 + ... + dp3N = 0.
∂q1 ∂q3N ∂p1 ∂p3N
But this expression is just
dH = 0.
Then, we have a characteristic curve given by
C = H(q, p),
and the general solution of Eq. (3.76) is given by any arbitrary function of
this characteristic
ρ = ρ(H). (3.78)
An usual ensemble of particle is the so called “canonical ensemble” defined

by
ρ = Z −1 e−βH , (3.79)
where Z −1 is a normalization constant called “partition function”, deduced
from
Z 3N
Y
ρ(q, p) dpi dqi = 1 (3.80)
i=1
and is given by
Z 3N
Y
Z= e−βH dpi dqi . (3.81)
i=1
The factor β is given in terms of the temperature (T ) and the Boltzmann’s

constant (k) as β = 1/kT . The above expression must have the correction
factor h3N N ! coming from the Quantum Statical Mechanics. The factor
N ! takes into account the degenerations due to having undistinguished par-
ticles, and the factor h3N is necessary in order for Eq. (3.81) to have the
right units and represents the minimum volume which can be occupied in
the phase space by the N particles. Thus, Eq. (3.79) and Eq. (3.81) are
written as
1 e−βH
ρ(q, p) = (3.82a)
N !h3N Z
and
Z 3N
1 −βH
Y
Z= e dpi dqi . (3.82b)
N !h3N i=1
The acknowledge of the partition function (3.82b) allows us know the

thermodynamics properties of many particle dynamical system, see refer-
ence K. Huang.
Example 3.5. For the example 6 given in first chapter, integrate Eq. (1.89)
and corroborate the result (3.78).
Eq. (1.89) is given as
∂ρ αp ∂ρ
+ = 0. (3.83)
∂q m ∂p
The equations for its characteristics are
dp dρ
dq = = .
(α/m)p 0
3.5. Renormalization Group’s Equations 73
From the first two terms, we get the characteristic curve

C = pe−αq/m ,
and the Hamiltonian is given in terms of this characteristic curve Eq. (1.86)
as
1 2 p2 −2αq/m
H(q, p) = C = e .
2m 2m
Then, the solution of Eq. (3.83) can be written as
2
p
ρ(q, p) = G e−2αq/m ,
2m
where G is an arbitrary function.
3.5 Renormalization Group’s Equations

If any field theory, a bare vertex is inevitably renormalized by higher order
correction diagrams (see Fig. 3.2), and these depend on the momentum of
the particle coming into the vertex.
+ + +· · ·
Fig. 3.2 Pair creation and some higher orders diagrams.
This means that we must introduce an effective momentum-dependent

vertex,
−iΥµ Tij g(µ)|p2i = −µ2 . (3.84)
In calculating higher order diagrams, we generally encounter divergences,
for example, in the scalar field theory the loop shown in Fig. 3.3 is loga-
rithmically divergent, as the loop momentum k goes to infinity. This fact
can be seen from the integral associated to this loop,
dk 4 λ2
Z
. (3.85)
(2π) (p/2 + k)2 (p/2 − k)2
4

Fig. 3.3 fig15 Scalar loop.
One way the renormalize the theory (i.e. to the eliminate all the infinite) is
introducing a momentum scale parameter, which depends on choice. Differ-
ent parameterizations are related by the renormalization group equations
dΓnB nF
µ = −(nB ΥB + nF ΥF )ΓR nB nF (pi , g, µ), (3.86)
dµ
where ΓRnB nF (pi , g, µ) is the renormalized Green function which is related
to the cut-off Green function ΓU nB nF (pi , g0 , Λ) as
nB nF
ΓR
nB nF (pi , g, µ) = [ZB (Λ/µ)] [ZF (Λ/µ)] ΓU
nB nF (pi , g0 , Λ), (3.87)
where ZB and ZF are dimensionless factors, nB and nF can be 0,1 or 2,
Λ is the cut-off momentum in all the integrations (like Eq. (3.85)) and is
introduced by hand to obtain finite quantities as ΓU nB nF (pi , g0 , Λ), g0 is
called the bare gauge coupling. The renormalized coupling constant g is
given by
g(µ) = ZB ZF ΓU
1,2 . (3.88)
ΥB and ΥF are defined as
∂ log ZB
ΥB (µ) = − lim µ
∆→∞ ∂µ
and
∂ log ZF
ΥF (µ) = − lim µ .
∆→∞ ∂µ
The Green function expresses the so called “propagator ” between two ver-
texes of a Feynman’s diagram. The effect of a change in the scale of external
momenta in the renormalized Green functions is expressed by
ΓR = ΓR D 2 2
nB nF (λ, pi , g, µ) = µ f (λ pi pj /µ ), (3.89a)
or equivalently by the scaling equation

∂ ∂
λ +µ ΓR = DΓR , (3.89b)
∂λ ∂µ
3.5. Renormalization Group’s Equations 75
where D is a constant which tell us the degree of homogeneity. Combining

the scaling equation (3.89b) and the renormalization group equation (3.86),
we obtain the following PDEFO
∂ΓR ∂ΓR
−λ + β(g) = −∆(g)ΓR , (3.90)
∂λ ∂g
where ∆(g) and β(g) are given by
∆(g) = D + nB ΥB (g) + nF ΥF (g) (3.91a)
and
∂g(µ)
β(g) = µ(g ). (3.91b)
∂µ
In addition, the following condition
ΓR (1, g) = GR (pi , g), (3.91c)
must be satisfied since for λ = 1 it is known the expression for the coupling
constant and the Green function. The characteristic curves of Eq. (3.90)
are found solving the equations
dλ dg dΓR
= = . (3.92)
−λ β(g) −∆(g)ΓR
From the first two terms, we get the characteristic curve given by
Z
dg
C = λ exp , (3.93)
β(g)
and from the last two terms, one can obtain
Z
∆(g)
ΓR = K(C) exp − dg (3.94)
β(g)
or
ΓR (λ, g) = K(λψ(g))φ(g), (3.95)
where ψ(g) and φ(g) are given by
Z
dg
ψ(g) = exp (3.96a)
β(g)
and
Z
∆(g)
φ(g) = exp − dg (3.96b)
β(g)
K(C) is an arbitrary function of the characteristic curve (3.93) and is de-
termined by the condition (3.91c). Using this condition in Eq. (3.95), we
have
GR (pi , g) = K(λψ(g))φ(g)
that is
GR pi , ψ −1 (s)

K(s) =
φ (ψ −1 (s))
or
GR pi , ψ −1 (λψ(g))

K(λψ(g)) = .
φ (ψ −1 (λψ(g)))
Using this expression in Eq. (3.95), it follows
φ(g)
ΓR (λ, g) = GR pi , ψ −1 (λψ(g))

. (3.97)
φ (ψ −1 (λψ(g)))
It is of particular interest for physicists to know what happens when λ → ∞.
According to Eq. (3.93) and Eq. (3.96a), we have
C
lim ψ(g) = lim = 0. (3.98a)
λ→∞ λ→∞ λ
Then, this implies from Eq. (3.96a) that
Z
dg
lim → −∞, (3.98b)
λ→∞ β(g)
for any g. In particular this must happen for g = 0. Therefore, Eq. (3.98b)
tell us that the function β(g) can be of the form
β(g) = an g n , (3.99)
where an is a constant and n can be 1, 2, 3, .... This happens for example to
the diagrams of Fig. 3.4, where β(g) is expressed as β(g)E − g 3 + O(> 3).
From left to right in this figure, the first loop corresponds to gluons, the
second one to ghosts, and the last one to quarks. This expresses what
is called “asymptotic freedom” for Quantum Cromo Dynamics (QCD). A
complete discussion about this can be found in reference of J.Ellis and C.T.
Sanchrajda.
3.6 Particle Multiplicity Distribution in High Energy

Physics
Collisions of two beam of particles (proton (p)-antiproton (p̄); electron (e− )-

positron (e+ ); etc.) at high energies, from 10 GeV up to around 900 GeV
(1GeV= 109 eV) produces more new charged (±) or neutral (o) particles
P±,o
such as pions (π ±,o ), Kaons (K ± ), Lambdas (Λo ), Sigmas ( ), etc.,
which are called “secondary particles”. The amount of secondary particles

3.6. Particle Multiplicity Distribution in High Energy Physics 77
+ +
Fig. 3.4
produced is called the “multiplicity of the event”. If our detector is capable

of knowing all the particles appearing after the collision, one talks about
”exclusive events detector”. If our detector can only know with detail few of
these outgoing particles, one talks about of an ”inclusive events detector.”
One is interested in to find, for a given energy collision, the number of
charged particles (or neutral particles) which can appear after the collision,
i.e. the multiplicity. Looking at the problem in a different way, we want to
find the probability Pn of finding a number “n” of charged particles for a
given energy. A possible probability distribution is given in the reference
of P. Srivastava,
Bn
Pn = e−<n> An (k exp(−B)), (3.100)
n!
where < n > denote the average number of charged particles, k is referred
to the k − th moment < xk >, B is defined as B =< n > /k, and An (x)
are the polynomials defined by the recurrent relation

dAn (x)
An+1 = x An (x) + (3.101)
dx
satisfying the initial condition
A0 (x) = 1. (3.102)
One approximation that we can make to Eq. (3.100) is by assuming that
the variable “n“ changes continuously. Thus, the term of Eq. (3.101) can
be taken at first order approximation in the Taylor’s series as
∂An (x)
An+1 (x) = An (x) + . (3.103)
∂n
In his way, making use of Eq. (3.103) and Eq. (3.101), and after rearranging
terms, a PDEFO is obtained
∂A ∂A
−x = (x − 1)A, (3.104)
∂n ∂x
where we have defined A = A(n, x) = An (x). Eq. (3.102) is

A(0, x) = 1. (3.105)
The equation for the characteristic of Eq. (3.104) are given by
dn dx dA
= = . (3.106)
1 −x (x − 1)A
From the first two term, we obtain the equation
dx
dn = (3.107)
−x
which can be integrated, obtaining the characteristic curve
C1 = n + log x. (3.108)
The characteristic curve can be chosen as C = exp(C1 ), that is
C = xen . (3.109)
From the last two terms of Eq. (3.106), we have the equation
dx A
= (3.110)
−x (x − 1)A
which can be written as
(x − 1)dx
Z Z
dA
= (3.111)
x A
and can be integrated, obtaining the solution
A = K(C)xe−x , (3.112)
or using Eq. (3.109), it follows that
A(n, x) = K(xen )xe−x . (3.113)
with the condition (3.105) in Eq. (3.113), we have
1 = K(x)xe−x,
or
K(s) = s−1 es .
Then, the function K is determined as
n
K(xen ) = x−1 en exe , (3.114)
and the solution of Eq. (3.104) is finally given as
A(n, x) = e−(x+n) exp(x exp(n)). (3.115)
3.7. Hamiltonian Perturbation Approach in Accelerator Physics 79
3.7 Hamiltonian Perturbation Approach in Accelerator

Physics
Accelerator Physics studies the behavior of a charged particle inside an ac-

celerator machine. Although there are many accelerator structures, the one
of particular interest for applications is the synchrotron machine where the
charged particles travel inside a ring, and the trajectories are determined
mainly by the magnetic field, B, provided by the ideal multipole magnets
placed all over the ring (dipole magnets are use for bending, quadrupole
magnets are use for focusing, sextupole magnets are use for chromatic-
ity corrections, etc.). These ideal (not constructions errors are assumed)
magnets define the so called “lattice” of the synchrotron machine. The
transverse behavior of the charged particle in this structure can be de-
scribed using the Hamiltonian approach where the Hamiltonian is given in
reference E.D. Courant and H.S. Snyder, and the reference M. Sand,
2 2
1 e 1 e
H= Px − A x + Py − A y
2 cp 2 cp

e 1 1 1
− As + K1 (s)y 2 − K1 (s) − 2 x2 , (3.116)
cp 2 2 ρ
where p is the longitudinal linear momentum of the charged particle,
Px = px /p and Py = py /p are its normalized transversal momenta, e is
the charged of the particle, c is the speed of light, ρ(s) it the curvature of
the accelerator ring, K1 (s) describes the linear lattice (ideal dipoles and
quadrupoles) of the machine, and the vector potential, A = (Ax , Ay , As ),
represents nonlinear magnets or additional electromagnetic field due to
other possible sources (nonlinear components of the real linear magnets,
or longitudinal multipole periodic pattern along the real magnets, for ex-
ample). Separating the nonlinear term of the vector potential, this Hamil-
tonian can be written as
H(x, P, s) = H0 (x, P, s) + V (x, P, s), (3.117)
where x and P are the vectors x = (x, y) and P = (Px , Py ), and H0 and V
are given by
1 2 1
H0 (x, P, s) = (P + Kx (s)x2 ) + (Py2 + Ky (s)y 2 ) (3.118a)
2 x 2
and
e
V (x, P, s) = − (As + Px Ax + Py Ay ). (3.118b)
cp
Note that the description is done taking the length of the accelerator ring
“s” as the independent parameter. It is convenient to make an action-angle
variables canonical transformation through the generating function
2
X x2i
F (x, φ, s) = − (tan(φi ) − βi0 (s)/2), for i = 1, 2 (3.119)
i=1
2βi (s)
where x is for i = 1 and y is for i = 2, βi0 (s) is the derivative with respect
to s of the beta function βi (s) which satisfies the differential equation
2βi βi00 − (βi0 )2 + 4Ki (s)βi2 = 4, (3.120a)
and the φi (s) is called the betatron phase which is related with the beta
function through
Z s
dσ
φi (s) = φi (0) + . (3.120b)
0 βi (σ)
The action, Ii , the coordinates, and the canonical linear momenta are given
by
∂F 1
Ii = − = [x2 + (βi x0i − βi0 xi /2)], (3.121a)
∂φi 2βi (s) i
p
xi = 2Ii βi cos(φi ), (3.121b)
and
s
2Ii 1 0
Pi = − sin(φi ) − βi cos(φi ) , for i = 1, 2. (3.121c)
βi 2
Therefore, Hamiltonian (3.117) written in terms of the action-angle vari-
ables (φ, I) is given by
2
X Ii
H(φ, I, s) = + V(φ, I, s), (3.122)
β
i=1 i
(s)
where the function V is defined as
V(φ, I, s) = V (x(φ, I), P(φ, I), s). (3.123)
One the most important parameters that the accelerator physicists are in-
terested in is the so called “tune” of the machine,
Z C
1 dσ
νi = , for i = 1, 2, (3.124)
2π 0 βi (σ)
which has the meaning of number of betatron oscillations performed by the
particle around the ring of length C. If the function V in Eq. (3.122) is dif-
ferent from zero, this tune is modified (the tune is shifted if the modification
is the same for all the particles, or there is “tune spread” if the modification
is different for different particles). To calculate this change at first order
in perturbation theory, the averages over the phases (φi ∈ [0, 2π], i = 1, 2)
and the integration over the length of the ring on the Hamiltonian (3.122)
is carried out
Z C Z 2π Z 2π
1
H(I)
e =< H(φ, I, s) >= ds dφi dφ2 H(φ, I, s). (3.125)
(2π)3 0 0 0
And the modified tune (νi0 ) is just the derivative of this averaged Hamilto-
nian with respect the action
∂ H(I)
e
νi0 = . (3.126)
∂Ii
To go to a second-order perturbation, a new canonical transformation close
to the identity can be made. This is achieved with the help of the generating
function
2
X
Fe(φ, J, s) = φi Ji + G(φ, J, s). (3.127)
i=1
The action between the new variables (Φ, J) and the old one (φ, I) is given
by the expressions
∂ Fe ∂G
Ii = = Ji + (3.128a)
∂φi ∂φi
and
∂ Fe ∂G
Φi = = φi + . (3.128b)
∂Ji ∂Ji
Hamiltonian (3.122) written in terms of these new variables is
2 2
X Ji X 1 ∂G ∂G e
H(Φ, J, s) = + + + V(Φ, J, s), (3.129)
β (s) i=1 βi (s) ∂φi
i=1 i
∂s
where the function V

e is given by

∂G ∂G
V(Φ, J, s) = V Φ −
e ,J − ,s . (3.130)
∂J ∂φ
Doing a Taylor expansions of this function
2
X ∂V ∂G ∂V ∂G
V(Φ,
e J, s) = V(Φ, J, s) + − + ··· , (3.131)
i=1
∂Ji ∂Φi ∂Φi ∂Ji
and neglecting terms of order higher than one, the Hamiltonian would be
written as
2 2
X Ji X ∂V ∂G ∂V ∂G
H(Φ, J, s) = + − , (3.132)
β (s) i=1 ∂Ji ∂Φi
i=1 i
∂Φi ∂Ji
if the function G satisfies the following linear partial differential equation
2
X 1 ∂G ∂G
+ + V(Φ, J, s) = 0, (3.133)
i=1 i
β (s) ∂φi ∂s
where it has been assumed φ ≈ Φ. The equation for the characteristics are
given by
dG
β1 (s)dφ1 = β2 (s)dφ2 = = ds. (3.134)
−V(φ, J, s)
From the first two terms and the last one of Eq. (3.134), two characteristic
curves follow
C1 = φ1 − ψ1 (s) and C2 = φ2 − ψ2 (s), (3.135)
where the function ψi (s) is defined as
Z s
dσ
ψi (s) = for i = 1, 2. (3.136)
0 βi (σ)
Using these equations in third term of Eq. (3.134), it follows that
Z s
G(φ, J, s) = − V(C1 + ψ1 (ξ), C2 + ψ2 (ξ), J, ξ)dξ + A(C1 , C2 ). (3.137)
0
Choosing the conditions G(φ, J, 0) = 0, the solution of Eq. (3.133) is given
by
Z s
G(φ, J, s) = − V(φ1 −ψ1 (s)+ψ1 (ξ), φ2 −ψ2 (s)+ψ2 (ξ), J, ξ)dξ. (3.138)
0
This solution is, then, substituted in Eq. (3.132) to be able to make the
average and calculate the modified tune of the machine through Eq. (3.126)
(for a explicit application see G. López and S. Chen). Assume the pertur-
bation function is periodic on the angles and distance variables. Then, this
function can be written in Fourier expansion as
(J) ei(m1 φ1 + m2 φ2 − lΩs) .
X
V(φ, J, s) = V m1 ,m2 , l
m1 ,m2 , l
Therefore, Eq. (3.138) can be expressed as

X
V φ1 − ψ1 (s) + ψ1 (ξ), φ2 − ψ2 (s) + ψ2 (ξ), J, ξ = Vm1 ,m2 , l (J)
m1 ,m2 , l
× ei[m1 (φ1 − ψ1 (s) + ψ1 (ξ)) + m2 (φ2 − ψ2 (s) + ψ2 (ξ)) − lΩξ] , (3.139)

and substituting this expression in Eq. (3.138), we obtain
Vm1 ,m2 , l (J)ei[m1 (φ1 − ψ1 (s)) + m2 (φ2 − ψ2 (s))]

X
G(φ, J, s) =
m1 ,m2 , l
Z s
× ei[m1 ψ1 (ξ) + m2 ψ2 (ξ) − lΩξ] dξ. (3.140)
0
Suppose now that ψ(s) ≈ ωi s, the integral of the right hand of (3.140) can
be expressed as
Z s Z s
ei[m1 ψ1 (ξ) + m2 ψ2 (ξ) − lΩξ] dξ = ei[m1 ω1 + m2 ω2 − lΩ]ξ dξ
0 0

 1 − ei[m1 ω1 + m2 ω2 − lΩ]s if m1 w1 + m2 w2 6= lΩ,
i

m1 ω1 + m2 ω2 − lΩ

= (3.141)



s if m1 w1 + m2 w2 = lΩ.
The first case (m1 ω1 + m2 ω2 6= lΩ) is called non resonant case, and the
other one (m1 ω1 + m2 ω2 = lΩ) is called resonant case. In this latter sit-
uation one says that one has non linear resonances characterized by the
pair (m1 , m2 ). Because in the non linear resonant situation the function
G grows linearly with the parameter s, there will be a value s∗ for which
this approximation will not be valid any more since it will have a big value.
These non linear resonance are import, for example, for determine the life
time of a beam in a circular accelerator, the stability of a planetary system,
the chaotic behavior of a diatomic molecule. Given a non linear resonance
(m1 , m2 ), this defines a rational relation among the involved frequencies,
ω1 = − m l
m1 ω2 + m1 Ω, and since the rational numbers are dense in the real
2
number, any operational point in the space (ω1op , ω2op ) of the system will
be close enough of a non linear resonance which could make it unstable.
Of course, nobody is interested in having a non linear system which last
forever. So, one just try to operate the system away of the ” danger non
linearities ” for the machine to have the life time required. For natural
periodic systems, one first chooses their operation point, then one tries to
determine all non linearities, and finally one tries to make some estimation
of their stability lifetime.
The procedure outlined in this section is called Kolmogorov perturba-

tion approach (see reference V.I. Kolmogorov), and note from Eq. (3.137)
that the function G is proportional to the perturbation V(Φ, J, s). There-

fore, Eq. (3.131) represents the Hamiltonian approximated up to second
order in V. If we call the second term on the right side of Eq. (3.131) as
V (2) (Φ, J, s), the resulting expression will have exactly the same form as
Eq. (3.121). Then, one can do another infinitesimal canonical transfor-
mation with a similar generatrix function (3.126) in terms of the variables
{Φ, J, s}. So, following the same procedure we have done before with this
Hamiltonian, we gets a similar expressions (3.131) and (3.137) but with
function V (2) instead of V, that is, Eq. (3.137) would be of second order
in V, and Eq. (3.131) would be of fourth order in V. Thus, in general,
for a nth-procedure, one gets and approximation of 2n -order in V with a
resulting Hamiltonian
2
X Ji n
H (n) (Φ, J, s) = + V (2 ) (Φ, J, s) ,
β (s)
i=1 i
n
where V (2 )
is given by
2 n−1 n−1
!
(2n )
X ∂V (2 )
∂G ∂V (2 )
∂G
V (Φ, J, s) = −
i=1
∂Ji ∂Φi ∂Φi ∂Ji
and the function G has the following expression
Z s
n−1
V (2 ) Φ1 − ψ1 (s) + ψ1 (ξ), Φ2 − ψ2 (s) + ψ2 (ξ), J, s dξ .

G(Φ, J, s) = −
0
Of course, one must have that the following relations must be satisfied
2 Ji
X
(2n ) (2n−1 )
V < V , and V
βi (s)
i=1
+
for Φi ∈ [o, 2π] , s, Ji ∈ [0, ∞) , n ∈ Z . On each procedure there will be a
modification on the tune through the expression
∂hH (n) i
(n)
νi =
i = 1, 2 ,
∂Ji
with the average defined as Eq. (3.124). If the procedures converge, it does
in an exponential way, 2n , see also reference V.I. Arnold.
3.8 Perturbation Approach for the One-dimensional Con-

stant of Motion
In what follows, one is restricted oneself to one-dimensional system for sim-

plicity. As it was shown in first section of this chapter, the constant of
3.8. Perturbation Approach for the One-dimensional Constant of Motion 85
motion of an autonomous system always exists and is given by the solution

Eq. (3.3). However, in most cases this equation is quite difficult to solve
(specially for nonlinear forces). Due to the importance of knowing this con-
stant of motion, one would like to know how this one is, at least, with some
degree of approximation. To do this, the following perturbation method
may be used. Suppose that the force can be separated in two parts,
f (q, v) = f0 (q, v) + fI (q, v). (3.142)
The first part, f0 , is the part of the force for which Eq. (3.3) is integrable
and has the known solution K0 (q, v),
∂K0 ∂K0
v + f0 (q, v) = 0,
∂q ∂v
and the second part, fI , is considered as a perturbation to the system.
Suppose also that the constant of motion can be expressed of the form
X∞
K(q, v) = K0 (q, v) + K (n) (q, v), (3.143)
n=1
(n)

where K is of the order (fI ) in force strength, that is O K (n) ∼ (fI )n .
n
Defining K (0) = K0 and substituting this expressions in

∂K ∂K
v + [f0 (q, v) + fI (q, v)] = 0, (3.144)
∂q ∂v
it follows
∞
∂K (n)

∂K0 ∂K0 X
v + f0 (q, v) + fI (q, v) +
∂q ∂v n=1
∂v
∞
∂K (n) ∂K (n) ∂K (n−1)
X
v + f0 (q, v) + fI (q, v) = 0. (3.145)
n=1
∂q ∂v ∂v
Neglecting the terms of the order (fI )n+1 , the following equation can be
derived
∂K (n) ∂K (n) ∂K (n−1)
v + f0 (q, v) = −fI (q, v) , for n = 1, 2... (3.146)
∂q ∂v ∂v
This is a linear partial differential equation of first order, and its equations
for the characteristics are
dq dv dK (n)
= = (n−1)
, (3.147)
v f0 (q, v) −fI (q, v)Kv
where K (n−1) means the partial differentiation of K (n−1) with respect to
v. From the first two terms of this equation one can obtain the same
function K0 as a characteristic curve of Eq. (3.146) which can be used to

get v = v(K0 , q) which, in turn, is used in Eq. (3.147) to get
Z q (n−1)
(n) fI (ξ, v(K0 , ξ))Kv
K (q, v) = − dξ, (3.148)
v(K0 , ξ)
or applying iteratively this equation, the constant of motion for the au-
tonomous system is given by
∞
" n Z ξi−1
#
X
n
Y fI (ξi ) ∂
K(q, v) = K0 (q, v) + (−1) dξi K0 , (3.149)
n=1 i=1
v(K0 , ξi ) ∂v
where ξ0 = q (for application to explicit examples see reference N.M. Atak-
ishiyev et al).
3.9 Constant of Motion for a Relativistic Particle under

Periodic Perturbation
The one-dimensional motion of the relativistic particle under the periodic

perturbation force can be described by the following Newton equation of
motion
3/2
d2 x v2

m 2 =e 1− 2 E sin (kx − ωt), (3.150)
dt c
where e, m and v are the charged, the mass at rest and the velocity of the
particle. E, k and ω are the amplitude, wave number and frequency of
oscillation of the field, c is the speed of light. Defining the new variables ξ
ξ = kx − ωt (3.151)
Eq. (3.150) can be written as the following autonomous dynamical system
dξ
= kv − ω, (3.152a)
dτ
3/2
v2

dv
= 1− 2 A sin ξ, (3.152b)
dτ c
where the coefficient A is defined as A = eE/m. A constant of motion of
this system is given by a solution of the following equation
3/2
v2

∂K ∂K
(kv − ω) + 1− 2 A sin ξ = 0. (3.153)
∂ξ c ∂v
The equation for the characteristic curves are
dξ dv dK
= 3/2
= . (3.154)
kv − ω 1− v
2
A sin ξ 0
c2
3.10. Uniqueness of constant of motion 87
Solving these equations, the constant of motion which the right limit for
the non-relativistic case (c → ∞) is given by
mc2 mωv mA
K(ξ, v) = p − mc2 − p + cos(kx − ωt). (3.155)
2
1 − v /c2 2
k 1 − v /c2 k
3.10 Uniqueness of constant of motion
In section (3.1), a criteria for getting a constant of motion for a dissipative

system was given. This criteria is based on taking the limit for the dissi-
pative parameter to zero, and looking for the functionality of the arbitrary
function such that one gets the know function at this limit (the usual energy
of a conservative system). However, this criteria does not guarantee that
this constant of motion is unique, in fact, consider the following dynamical
quadratic dissipative system under a constant force mf , being m the mass
of the particle,
dx dv α
=v , = f − v2 , (3.156)
dt dt m
where α is the dissipation parameter, and one has v ≥ 0. The constant of
motion of this autonomous system K(x, v) satisfies the linear PDEFO
∂K α ∂K
v + f− v =0 (3.157)
∂x m ∂v
which has the general solution
αv 2

mf
K(x, v) = G C(x, v) , with C(x, v) = ln 1 − − fx ,
2α mf
(3.158)
where G is an arbitrary function of the characteristic curve ”C”. By choos-
ing
m2 f −2αC/m m2 f
G(C) = mC , or G(C) = − e + (3.159)
2α 2α
one gets the constants of motion
m2 f αv 2

(1)
Kα (x, v) = − ln 1 − − mf x , (3.160)
2α mf
or
m2 mv 2 m2 f

(2)
Kα (x, v) = − f− e2αx/m + . (3.161)
2α m 2α
Using Taylor expansion on both constants of motion, one has that
1
lim Kα(i) (x, v) = mv 2 − mf x , for i = 1, 2 . (3.162)
α→0 2
Both different constants of motion has the same expected ”physical” limit,
and this represents a strong ambiguity for Lagrangian, Hamiltonian and
applications on quantum and statistical mechanics.
To overcome this ambiguity when dealing with velocity dependent forces,

let us recall from chapter 2 that the theorem of existent and uniqueness of
a linear and quasilinear PDEFO is stablished whenever the curve of initial
data is not one of the characteristic curves. Therefore, assuming that the
curve of initial data

1 2
Γ0 : x = 0 , K = mv (3.163)
2
is not a characteristic curve of the linear PDEFO
∂K F (x, v) ∂K
v + =0, (3.164)
∂x m ∂v
associated to the constant of motion of the dynamical system
dx dv F (x, v)
=v , = , (3.165)
dt dt m
the solution will be unique determined (without the ambiguity mentioned
before).
Let us apply this idea to the above dynamical system. The general so-
lution is given by Eq. (3.157), and applying the initial condition of Eq.
(3.162), K(0, v) = mv 2 /2, it follows that
αv 2

1 mf
mv 2 = G − ln (1 − ) (3.166)
2 2α mf
Defining σ as
αv 2

mf
σ=− ln 1 − , (3.167)
2α mf
one has that
mf
v2 = 1 − e−2ασ/mf (3.168)
α
and
m2 f
G(σ) = 1 − e−2ασ/mf . (3.169)
2α
3.10. Uniqueness of constant of motion 89
Therefore and after some rearrangements, the constant of motion which

pass on the above initial curve is given by
m2 f αv 2

Kα (x, v) = 1− 1− e2αx/m . (3.170)
2α mf
This constant of motion has the expected limit
1
lim Kα (x, v) = mv 2 − mf x . (3.171)
α→0 2
This curve of initial conditions looks to work fine for the non relativistic
case (v c, with ”c” being the speed of light). However, for the relativistic
case (v / c) it is better to select the following curve of initial data
( −1/2 )
v2

2
Γ0 : x = 0 ; K = (γ − 1)mc |where γ = 1 − 2 . (3.172)
c
The reason for doing this is that for a free relativistic particle one knows
that the energy is given by (γ − 1)mc2 , having the right non relativistic
limit when v/c 1 (that is, mv 2 /2).
Consider a relativistic particle moving under a conservative force F (x).

Its associated dynamical system can be written as
3/2
v2

dx dv F (x)
=v , = 1− 2 (3.173)
dt dt m c
The constant of motion (K(x, v)) for this autonomous system is the solution
of the following linear PDEFO
3/2
v2

∂K F (x) ∂K
v + 1− 2 = 0. (3.174)
∂x m c ∂v
Its characteristic equations are given by
dx dv dK
= = (3.175)
v F (x) v 2 3/2 0

m 1− c2
which has the general solution

K(x, v) = G C(x, v) , (3.176)
where ”C” is the characteristic curve deduced form the first two term of this
expression. Separating variables, one arrives to the following integration
Z Z
vdv
F (x)dx = m 2 3/2
, (3.177)
1 − vc2
and its integration brings about the characteristic curve

C(x, v) = γmc2 + V (x) , (3.178)
R
where V (x) represents the potential energy (V (x) = − F (x)dx), and γ
is the known function γ(v) = (1 − v 2 /c2 )−1/2 . Let us call V0 = V (0) and
apply the initial data to our solution, obtaining the expression
(γ − 1)mc2 = G(γmc2 + V0 ) . (3.179)
Defining now σ = γmc2 + V0 , one gets
G(σ) = σ − V0 − mc2 . (3.180)
Having this functionality already, after making the evaluation of G in
γmc2 + V (x), one obtains the solution
K(x, v) = (γ − 1)mc2 + V (x) − V0 . (3.181)
This constant of motion is the usual relativistic expression for a conser-
vative system with one degree of freedom, has the right limit for the non
relativistic case, and is uniquely defined.
Of course, it is not always possible to get v = v(σ) during the process

of the determination of the function G, for example, consider a relativistic
quadratic dissipative system (v ≥ 0) under a constant force F , described
by the dynamical system
3/2
αv 2 v2

dx dv F
=v , = − 1− 2 . (3.182)
dt dt m m c
The constant of motion for this autonomous system is the solution of the
following liner PDEFO
3/2
αv 2 v2

∂K F ∂K
v + − 1− 2 =0 (3.183)
∂x m m c ∂v
which is given by K(x, v) = G(C(x, v)), where C(x, v) is the characteristic
curve obtained from the integration of the first two therm of the following
equations for the characteristics
dx mdv dK
= 2 2 2 3/2
= (3.184)
v (F − αv )(1 − v /c ) 0
which is given by
√ r
γ aα 1 aα
C(x, v) = mc2 + arctan − Fx ,
1 − aα (1 − aα )3/2 γ 1 − aα
(3.185)
3.11. Vlasov Equation and Bunched Beam Instabilities 91
where aα has been defined as aα = αc2 /F . Note that this parameter has
the range 0 ≤ aα ≤ 1, and is singular for aα = 1. Note also that the
characteristic curve has the limit limα→0 C = γmc2 . Applying the initial
data to the solution, it follows that
√ r
γ aα 1 aα
(γ − 1)mc2 = G mc2 + arctan
1 − aα (1 − aα )3/2 γ 1 − aα
(3.186)
So, to find the functionality of G one proceeds as usual
√ r
γ aα 1 aα
σ = mc2 + arctan , (3.187)
1 − aα (1 − aα )3/2 γ 1 − aα
and from here it is clear
p that is not possible to get γ = γ(σ) in general but
for small values of aα /(1 − aα )/γ. For this particular weak dissipation
case, the above expression is given as
mc2

aα
σ= γ(1 − aα ) + (3.188)
(1 − aα )2 γ
which has the solution s 2
(1 − aα )σ (1 − aα )σ aα m2 c4
mc2 γ± = ± − . (3.189)
2 2 1 − aα
That is, even in this case the relationship between γ and σ corresponds to
two values function. In this way, G(σ) is given by
s 2
(1 − aα )σ (1 − aα )σ aα m2 c4
G(σ) = ± − − mc2 , (3.190)
2 2 1 − aα
and the constant of motion for this weak dissipation case would be given
by

1 − aα γ aα
K(x, v) = mc2 + − F x
2 1 − aα γ(1 − aα )2
s 2
mc2 aα m2 c4

aα (1 − aα )F x
± γ+ − − . (3.191)
2 γ(1 − aα ) 2 1 − aα
3.11 Vlasov Equation and Bunched Beam Instabilities
Consider an ensemble of N non interacting particles moving with n-degreed

of freedom where each particle obeys the same dynamical system defined
by the equation
dx
= F(x, t) , x ∈ <2n , t ∈ <, F = (f1 , . . . , f2n ) , (3.192)
dt
where ”t” is the parameter which describes the evolution of the system (non
autonomous system). The ensemble of N-particles is described by an scalar
function ψ(x, t) such that
Z
dψ
ψ(x, t)dx = N , and =0, (3.193)
<2n dt
that is, the function is normalized to the number of particles, and it is a
constant of motion where its explicit expression is
2n
X ∂ψ ∂ψ
fi (x, t) + =0. (3.194)
i=1
∂xi ∂t
There is not really a restriction for an even dimensional dynamical system
Eq. (3.192), one will take this case for the applications below. For the
particular case when Eq. (3.192) represents a Hamiltonian system, one has
that x = (q, p) with q, p ∈ <n and the dynamical system looks like
dq dp
= ∇q H , = −∇p H , (3.195)
dt dt
where H = H(q, p, t) is the Hamiltonian of the system, we will have the
function ψ = ψ(q, p, t), and the Vlasov equation will have the form
n
X ∂H ∂ψ ∂H ∂ψ ∂ψ
− + =0. (3.196)
j=1
∂pj ∂qj ∂qj ∂pj ∂t
If the system is autonomous (that is, F = F(x), or H = H(q, p), the term
∂ψ/∂t can be neglected on the Eq. (3.194) and Eq. (3.196).
3.11.1 Potential well distortion

Consider now a continuous beam of charged particles (a single bunch) mov-
ing inside a beam pipe of a circular accelerator (or stored ring) with a rev-
olution frequency of ωo = 2π/To (To is the revolution period) and with
an energy E. The relative error in the energy of the beam, δ = ∆E/E,
and the arrival time displacement of the beam at the accelerating cavity, τ ,
evolve with respect the longitudinal variable ”s” according to the following
dynamical system
dτ α dδ
=− δ , = g(τ ) , (3.197)
ds c ds
where α is a positive constant called the momentum compaction factor,
c represents the speed of light, and g(τ ) is a function which takes into
account the damping and the wake field effect (function W (τ ) generated in
a structure of length L) on the beam,
Z ∞
ω2 e2 L
g(τ ) = s τ − h(τ 0 )W (τ 0 − τ )dτ 0 , (3.198)
αc ETo c τ
where ”e” is the charge of the particles, ωs is the synchrotron oscillation
frequency (ωs ≤ ωo ), and h(τ ) is the reduced density part associated to the
variable τ ,
Z
h(τ ) = ψ(τ, δ) dδ . (3.199)
<
Our system is an autonomous Hamiltonian system with Hamiltonian given

by
Z τ
α
H(τ, δ) = g(τ 0 ) dτ 0 + δ 2 . (3.200)
0 2c
Thus, the Vlasov equation (n=1) for an ensemble of charged particles sat-
isfying this dynamical system is given by
∂ψ αδ ∂ψ
g(τ )− =0. (3.201)
∂δ c ∂τ
The equations for its characteristic curves are given by
dδ dτ dψ
= = , (3.202)
g(τ ) −αδ/c 0
where the Hamiltonian Eq. (3.200) is a characteristic curve, and the solu-
tion of the equation is
ψ(τ, δ) = G(H(τ, δ)) (3.203)
with G being an arbitrary function. Choosing at τ = 0 a Gaussian distri-

bution with standard deviation σ for ψ,
1 2 2
ψ(0, δ) = √ e−δ /2σ , (3.204)
σ 2π
it follows that the functionality G is determined by
1 2
G(ξ) = √ e−cξ/ασ . (3.205)
σ 2π
Therefore one gets the solution
1 2 2
ψ(τ, δ) = √ e−δ /2σ h(τ ) , (3.206)
σ 2π
where the normalization has taken equal to unit, and h(τ ) is the reduced
density defined as
Z τ
−c/ασ 2
h(τ ) = e g(τ 0 )dτ 0 . (3.207)
0
Substituting Eq. (3.198) in this expression, one gets an integral equation
for the determination of the reduced density
R∞
dτ 0 h(λ)W (λ−τ 0 ) dλ]
2 2
Rτ
/2α2 σ 2 −(e2 L/ασ 2 ETo )
h(τ ) = e−[ωs τ 0 τ0 (3.208)
which is normally solved by numerical methods. The solution brings about
the potential well distortion of an initial Gaussian bunch shape.
3.11.2 Longitudinal Modes

A more complicated situation is presented for a function g of the form
ω2 e2 ωo
g(τ, s) = s − V (τ, s) , (3.209)
αc To Ec
where V (τ, s) is defined as
+∞
X
V (τ, s) = e−iΩs/c ei(kωo +Ω)τ Z(kωo + Ω)ĥ(kωo + Ω) , (3.210)
k=−∞
being Z(ω) the longitudinal impedance of the accelerator (Fourier transfor-

mation of the wake field), ĥ(ω) is the Fourier transformation of the reduced
density, V (τ, s) is proportional to the voltage exited by the wake field, pro-
duced by h(τ ), at the location ”s” and seen by a particle at the point ”cτ ”.
We have assumed that there is a disturbance having a single frequency Ω
in the accelerator, and multi-turn wakes effect has been included. In this
case, the dynamical system is written as
dτ α dδ
=− δ , = g(τ, s) (3.211)
ds c ds
which is a non autonomous system, and the associated Vlasov equation is
given by
∂ψ αδ ∂ψ ∂ψ
g(τ, s) − + =0. (3.212)
∂δ c ∂τ ∂s
Using Eq. (3.209) in this expression, let us write it in the following form
ωs2 ∂ψ αδ ∂ψ ∂ψ e2 ωo ∂ψ
− + = V (τ, s) , (3.213)
αc ∂δ c ∂τ ∂s To Ec ∂δ
and solve this PDEFO through successive approximation method

(limn→∞ ψn = ψ), where the the nth-approximation is given by
ωs2 ∂ψn αδ ∂ψn ∂ψn
− + = An−1 (δ, τ, s) . (3.214)
αc ∂δ c ∂τ ∂s
The term An−1 has been defined as
e2 ωo ∂ψn−1
An−1 (δ, τ, s) = Vn−1 (τ, s) , (3.215)
To Ec ∂δ
with Vn−1 the corresponding term with the reduced density ĥn−1 , and with
ψo being the solution of the equation
ωs2 ∂ψo αδ ∂ψo
− =0. (3.216)
αc ∂δ c ∂τ
Since the operator associated to Eq. (3.212) satisfies the Lipschitz condition
on a compact set of the variables τ , δ, and s, and it there is not singularities
of any ψn on this set, the succession converge to the solution of the equation
(given the initial value). So, let us assume that his happens and solve the
problem at first approximation. The solution at order zero was already
done, and it is given by Eq. (3.206), in turns, this implies that ĥ0 (ω) is
known. Then, the whole term A0 (δ, τ, s) is know. However, for the moment
let us assume that the term An−1 (δ, τ, s) is fully known and solve the general
problem. The equations for the characteristic of Eq. (3.214) are given by
dδ dτ dψn
= = ds = . (3.217)
ωs2 τ /αc −αδ/c An−1 (δ, τ, s)
The first two terms of this expression brings about the characteristic curve
ω2 α
a1 = s τ 2 + δ 2 , (3.218)
2αc 2c
from which one gets
r
2ca1 ω2
δ=± − s2 τ 2 . (3.219)
α α
This expression can be substituted in the second term and together with
the third term, one obtains the second characteristic curve
 s 
ω 2 2αa1
s 
a2 = arcsin τ ± s. (3.220)
2cαa1 ωs
From these two characteristics one gets τ = τ (s) and δ = δ(s),

s r
2cαa1 2αa1 2ca1 2αa1
τ (s) = sin s + a2 , δ(s) = ± cos s + a2 .
ωs2 ωs α ωs
(3.221)
These characteristics are kept independently of the perturbation order.

Thus, one can use the last two term or Eq. (3.217) to get
Z s
ψn (δ, τ, s) = An−1 (δ(s0 ), τ (s0 ), s0 ) ds0 . (3.222)
0
So, the solution at order zero is given by Eq. (3.206),

1 2 2
ψ0 (δ, τ ) = √ e−δ /2σ h(τ ) , (3.223)
σ 2π
and using Eq. (3.215), at first order approximation the density of particles
is given by
Z s
e2 ωo
ψ1 (δ, τ, s) = − δ(s0 )V0 (τ (s0 ), s0 )ψ0 (δ(s0 ), τ (s0 )) ds0 (3.224)
To Ecσ 2 0
3.11.3 Transverse Modes

In this case, the beam has a dipole moment in the transversal plane (let us
choose the y-direction), and the dynamical equations of motion are given
by
dτ α dδ ω2 y ∂Fy
=− δ , = sτ + (3.225)
ds c ds αc cE ∂τ
and
dy dpy ωβ2 Fy
= py , =− 2y+ , (3.226)
ds ds c E
where Fy = Fy (τ, s) is the transversal wake force generated by the dipole
moment of the beam, ωβ is the unperturbed betatron (transversal) fre-
quency of the beam, and the meaning of other parameters is the same as
before. The associated Vlasov equation to the density of charged particles
has the following expression
!
ωβ2
2
ωs y ∂Fy ∂ψ αδ ∂ψ ∂ψ Fy ∂ψ ∂ψ
τ+ − + py + − 2y + = 0,
αc cE ∂τ ∂δ c ∂τ ∂y E c ∂py ∂s
(3.227)
with ψ = ψ(x, s), x = (δ, τ, y, py ) ∈ <4 , being the density function. As
before, let us write this equation in the form
!
ωs2 τ ∂ψ αδ ∂ψ ωβ2

∂ψ Fy ∂ψ ∂ψ y ∂Fy ∂ψ
− + py + − 2y + =− ,
αc ∂δ c ∂τ ∂y E c ∂py ∂s cE ∂τ ∂δ
(3.228)
and solve it through successive approximations, {ψn }, such that the nth-
approximation satisfies the PDEFO
!
ωs2 τ ∂ψn αδ ∂ψn ∂ψn Fy ωβ2 ∂ψn ∂ψn
− + py + − 2y + = An−1 (x, s),
αc ∂δ c ∂τ ∂y E c ∂py ∂s
(3.229)
where An−1 has been defined as

y ∂Fy ∂ψ
An−1 (x, s) = − . (3.230)
cE ∂τ ∂δ
The equations for its characteristics are given by
dδ dτ dy dpy dψn
= = = = ds = . (3.231)
ωs2 τ /αc −αδ/c py Fy /E − ωβ2 y/c2 An−1 (x, s)
From the first two terms, we can see that we can still have the same char-
acteristics as before,
ωs2 2 α
a1 = τ + δ2 (3.232)
2αc 2c
and
 s 
ωs2 

2αa2
a2 = arcsin τ ± s. (3.233)
2cαa1 ωs
Therefore, we can have the same s-dependence for τ and δ,

s
2cαa1 2αa1
τ (s) = sin s + a2 (3.234)
ωs2 ωs
r
2ca1 2αa1
δ(s) = ± cos s + a2 . (3.235)
α ωs
Now, third, fourth and fifth terms represent Eq. (3.226). So, making the
differentiation of y 0 , one gets
d2 y ωβ2
+ 2 y = Fy (τ (s), s)/E (3.236)
ds2 c
Z
ωβ ωβ s ωβ s ωβ s
y(s) = sin F̃ (s) cos ds + a3 sin
Ec c c c
Z
ωβ ωβ s ωβ s ωβ s
− cos F̃ (s) sin ds + a4 cos , (3.237)
Ec c c c
where the function F̃ has been defined as F̃ (s) = Fy (τ (s), s). Defining now
the functions f1 (s) and f2 (s) as
Z
ωβ ωβ s ωβ s
f1 (s) = sin F̃ (s) cos ds (3.238)
Ec c c
and
Z
ωβ ωβ s ωβ s
f2 (s) = cos F̃ (s) sin ds , (3.239)
Ec c c
Thus, the characteristics a3 and a4 have the following form
c n ωβ ωβ s ωβ s o
a3 = (y − f1 (s) + f2 (s)) sin + (py − f10 (s) − f20 (s)) cos
ωβ c c c
(3.240)
and
c n ωβ ωβ s ωβ s o
a4 = (y − f1 (s) + f2 (s)) cos − (py − f10 (s) − f20 (s)) sin ,
ωβ c c c
(3.241)
0 0
where f1 and f2 represent the differentiation of these function with resect
to ”s.” In addition, the explicit dependence of y and py on the variable ”s”
has the following form
ωβ s ωβ s
y(s) = f1 (s) − f2 (s) + a3 sin + a4 cos (3.242)
c c
and
ωβ h ωβ s ωβ s i
py (s) = f10 (s) − f20 (s) + a3 cos − a4 sin . (3.243)
c c c
In this way, from Eq. (3.230), one gets the function An−1 fully expressed
as a function of the parameter s,
Ãn−1 (s) = An−1 (x(s), s) , (3.244)
and the integration of the last two term of Eq. (3.229) can be done,
Z s
ψn (x, s) = Ãn−1 (s0 ) ds0 , (3.245)
0
or
Z s
1 0 ∂Fy ∂ψn−1
ψn (x, s) = − y(s ) ds0 (3.246)
Ec 0 ∂τ τ (s0 ),s0 ∂δ x(s0 ),s0
It is possible to solve the problem at any order of approximation by knowing

the zero order density function ψ0 . This zero order approximation satisfies
the following PDEFO
!
ωs2 τ ∂ψ0 αδ ∂ψ0 ∂ψ0 Fy ωβ2 ∂ψ0 ∂ψ0
− + py + − 2y + = 0. (3.247)
αc ∂δ c ∂τ ∂y E c ∂py ∂s
Since the equations for its characteristics are almost the same as before,
dδ dτ dy dpy dψ0
= = = 2 = ds = , (3.248)
ωs2 τ /αc −αδ/c py Fy /E − ωβ y/c2 0
its characteristics are the same (a1 , a2 , a3 , and a4 above), and the general
solution of this equation is given by
ψ0 (x, s) = G(a1 , a2 , a3 , a4 ) , (3.249)
where G is an arbitrary function of these characteristics. This functionality
is determined by the initial distribution of the particles. Suppose that at
s = 0 one has τ (0) = 0, py (0) = 0 and Fy (0, 0) = 0, and ψ0 (x(0), 0) is having
a Gaussian distribution in the variables δ and y with standard deviations σ
and σy . Then, from the above expression for the characteristics, it follows
that
1 2 2 1 2 2
ψ0 (x(0), 0) = G(αδ 2 /2c, 0, y, 0) = √ e−δ /2σ √ e−y /2σy (3.250)
σ 2π σy 2π
Thus, the functionality of G for any variables (ξ1 , ξ1 , ξ3 , ξ4 ) is written as
1 2 2 2
G(ξ1 , ξ2 , ξ3 , ξ4 ) = e−(cξ/2σ +ξ3 /2σy ) , (3.251)
2πσσy
and the solution is finally gotten as
1 2 2 2
ψ0 (x, s) = e−(ca1 /2σ +a3 /2σy ) , (3.252)
2πσσy
where a1 = a1 (δ, τ ) is given by Eq. (3.232), and a3 = a3 (y, py , s) is given
by Eq. (3.240). Since ∂ψ0 /∂δ = −(α/σ 2 )δψ0 , at first order approximation
the solution is given by the following integration
Z s !
α ∂ F̃ y
δ(s0 )y(s0 )ψ0 x(s0 ), s0 ds0 (3.253)

ψ1 (x, s) =
Ecσ 2 0 ∂τ 0 0 (x(s ),s )
3.12 Interaction Plasma-Electromagnetic Field
A plasma is a set of heavy charged particles called ions together with very
light charged particles called electrons in some compact region Ω ⊂ <3
such that outside some finite length (Debye length)1 from this region the
set looks as neutral. Plasma covers a large region of temperatures and
densities, solar wind, solar corona, earth ionosphere, flames, fusion reactors,
laser, plasma, and so on. One could say that there are two ways to treat
the plasma depending on whether or not this one is enough diluted since
this characteristic determine the interaction behavior among the elements
of the plasma.
3.12.1 Diluted Plasma

If there is not transport of ions, and the plasma is enough diluted such
that almost not collision occur among ions or electrons, this plasma can be
described by a distribution function f (x, v, t) associated to the electrons
which satisfies the Vlasov’s equation
∂f e h v i ∂f ∂f
v· + E+ ·B · + =0, (3.254)
∂x m c ∂v ∂t
where E and B are the electric and magnetic fields which are solutions of the
Maxwell’s equations (CGS units). For a isotropic-linear-homogeneous, non
dispersed, and local plasma with constant permittivity () and permeability
(µ) are given by
1 ∂B 4πµ µ ∂E
∇ · E = 4πρ/ , ∇×E = − , ∇·B = 0 , ∇×B =
J+
c ∂t c c ∂t
(3.255)
ρ and J are the charged and current densities which are given in terms of
the function f as
Z Z
ρ(x, t) = ρi − e f (x, v, t)dv , J(x, t) = Ji − e vf (x, v, t)dv
<3 <3
(3.256)
with ρi and Ji being the densities of charge and current ions, and it is
assumed that an equilibrium situation the distribution of electrons is sta-
tionary, depending only on the variable v, and it is of the type Maxwell
p
1 This Debye length is given by λD = kB T /4πe2 n, where kB is the Boltzmann con-
stant, T is the temperature, e and n are the electron charge and density, which
p can be
written as λD = 740 T /n cm with T given in eV and n in cm−3 , or λD = 6.65 T /n cm
p
when T is given in Kelvin degrees.

3.12. Interaction Plasma-Electromagnetic Field 101
distribution function at a temperature T ,

ρi p 2
f0 (v) = 2(βm)3 /π v 2 e−βmv /2 , (3.257)
e
satisfying
Z Z
e f0 (v)dv = ρi and e vf0 (v)dv = ρi v = Ji (3.258)
<3 <3
because non net charge or current must appear in the plasma (β = 1/kB T
with kB being the Boltzmann constant). The electric and magnetic fields
of Eq. 3.255 represent the outside field (homogeneous solution) and the
field produced by the charges (particular solutions). Eq. 3.254 represents
a non linear integro-partial-differential equation for the function f which is
the type of equation outside the topics of this book. However, we note that
the electromagnetic field is a linear operator acting on the function f , rep-
resented by the solution of the Eqs. 3.255 and 3.256. The solution is in fact
the retarded electromagnetic field (see reference J.D. Jackson and reference
J.R. Reitz), and let us call this vector operator F(x, v, t; f ), and note that
for our equilibrium distribution (if there is not external electromagnetic
field), one has F(x, v, t; f0 ) = 0,

e 1
F(x, v, t; f ) = E+ v×B . (3.259)
m c
In this way, one may propose an iterative method to solve this equation of
the form
∞
X
f (x, v, t) = fs (x, v, t) , (3.260)
s=0
such that f0 is our stationary Maxwell’s distribution function. Substituting

this expression in Eq. 3.254 and separating terms of the same order in fs ,
one gets the relation
s
∂fs ∂fs X ∂fs−l
v· + =− F x, v, t, fl · , s≥1 (3.261)
∂x ∂t ∂v
l=0
For s = 1 and using the above observation, one gets

∂f1 ∂f1 ∂f0
v· + = −F x, v, t, f1 · . (3.262)
∂x ∂t ∂v
For s = 2, one has
∂f2 ∂f2 ∂f0 ∂f1
v· + = −F x, v, t, f2 · − F x, v, t, f1 · . (3.263)
∂x ∂t ∂v ∂v
And so on. These type of equations are linear -integro-partial-differential

equations.
If we see Eq. 3.261 as a linear PDEFO in <7 , this equation would have the
following equations for their characteristics
dv1 dv2 dv3 dx1 dx2 dx3 fs
dt = = = = = = = Ps
0 0 0 v1 v2 v3 − l=0 F x, v, t, fl · ∂f∂v
s−l
(3.264)
which would have the characteristics
ci = vi , di = xi − vi t i = 1, 2, 3 (3.265)
independently of s approximation.
R Now, writing explicitly the dependence
on the type of operator, f (x, v, t)dv, from the first and last terms of Eq.
3.264, and using these characteristics, one has that the density distribution
fs comes from the solution of the following equation
s Z
dfs X ∂fs−l
+ F d + ct, c, t; fl (d + ct, v, t)dv · =0,
dt <3 ∂v (d+ct,c,t)
l=1
(3.266)
and given the initial conditions fj (0) for j = 1, . . . , s the unique solution of
this equation can be determined. Since fs ≥ 0 for x, v ∈ <3 and t ∈ <, the
Laplace transform
Z ∞
f˜s (p) = L{fs (t)} = e−pt f (t) dt,
0
being p a complex number, can be used to to solve this equation. Due to
L is a linear operation and has the properties
L{df /dt} = pL{f } + f (0) ,

R Rep+i∞
L{f · g} = (1/2πi) Re p−i∞
f˜(σ)g̃(p − σ)dσ ,
R R
L{ g(t)dt} = L{g}/p + (1/p)[ g(t)dt]t=0 , and
Rt
L{ 0
f (σ)g(t − σ)dσ} = L{f } · L{g}.
Applying this transformation in Eq. 3.266, one gets

s ^
e ? ∂fs−l = 0 ,
X
pf˜s + fs (0) + F (3.267)
∂v
l=1
Now, using the retarded fields in Eq. 3.259 and proceeding with Eq. 3.266
and Eq. 3.267 is one option to find fs . However, it is more convenient here
to use the representation of the equations in the Fourier space, where the
Fourier transformation
Z
ˆ 1
f (k) = F[f (x)] = eix·k f (x) dx (3.268)
(2π)3/2 <3
is a linear operation having the property F[∂f /∂xn ] = −ikn F[f ]. Using
this property in Eq. 3.255, it follows that
b =− 1 ∂B
b 4πµ µ ∂ E
b
ik· E
b = 4π ρ̂/, ik× E , ik· B
b =0, ik× B
b =
Ĵ+ .
c ∂t c c ∂t
(3.269)
Using the vector identity a × (b × c) = (a · c)b − (a · b)c in the second and
last terms and making some rearrangements, the following evolution equa-
tion are obtained in the Fourier space
∂2Eb
b = −i 4πcρ̂ k − 4π ∂ Ĵ
2
+ ω2 E (3.270)
∂t µ ∂t
∂2Bb
b = i 4πc k × Ĵ
+ ω2 B (3.271)
∂t2
√
where ω has been defined as ω = ck/ µ. Ignoring the homogeneous
solutions of these equations, their particular solutions are given by
Z t
4π t ∂ Ĵ(s)
Z
i4πc
Ep = −
b k ρ̂(s) sin ω(t − s)ds − sin ω(t − s)ds (3.272)
µω 0 ω 0 ∂s
Z t
b p = i4πc k ×
B Ĵ(s) sin ω(t − s)ds , (3.273)
ω 0
where the dependence only on time for the current and density has been
written for short writing. The Fourier transformation of the continuity
equation ∇ · J + ∂ρ/∂t = 0 brings about the relation
∂ ρ̂
ik · Ĵ + =0. (3.274)
∂t
Let us solve Eq. 3.262 using these tools and noting that ∂f0 /∂v depends
only on v. Applying the Fourier transformation to this equation, one gets
∂ fˆ1

e b 1 b · ∂f0
iv · kfˆ1 + =− E+ v×B (3.275)
∂t m c ∂v
or
∂ fˆ1 ∂f0 t
Z
ˆ i4πce
iv · k f1 + = k· ρ̂(s) sin ω(t − s)ds
∂t µωm ∂v 0
Z t
4πe ∂f0 ∂ Ĵ(s)
+ · sin ω(t − s)ds
ωm ∂v 0 ∂s
Z t
i4πce ∂f0
− v× k× Ĵ(s) sin ω(t − s)ds · (3.276)
ωm 0 ∂v
Using the same above vector identity, this equation can be written as
∂ fˆ1 ∂f0 t
Z
i4πce
iv · k fˆ1 + = k· ρ̂(s) sin ω(t − s)ds
∂t µωm ∂v 0
Z t
4πe ∂f0 ∂ Ĵ(s)
+ · sin ω(t − s)ds
ωm ∂v 0 ∂s
Z t
i4πce ∂f0
− k· v· Ĵ(s) sin ω(t − s)ds
ωm ∂v 0
Z t
i4πce ∂f0
+ v·k · Ĵ(s) sin ω(t − s)ds
ωm ∂v 0
(3.277)
Note that for the particular case when one considers Ĵ = 0, which is equiv-
alent to ignore the magnetic field contribution, the density ρ̂ must not
depend explicitly on time due to continuity equation ( Eq. 3.292), and the
above equation is reduced to the equation
∂ fˆ1

ˆ i4πce ∂f0
ik · v f1 + = v· ρ̂ , (3.278)
∂t µω 2 m ∂v
However from Eq. 3.256, one has an explicitly time dependence in the
density through the function f1 ,
Z
ρ̂(k, t) = −e fˆ1 (k, v, t)dv. (3.279)
<3
Therefore, to solve consistently Eq. 3.276, one needs to take into account
the dependence on time of the density. Ignoring this detail, the solution of
Eq. 3.278 leads to what is called ”Landau damping effect” (see reference
L. Landau and reference V. Maslo and M. Fedoryuk). With the above
properties of the Laplace transformation and knowing that L{sin ωt} =
ω/(p2 + ω 2 ), let us apply this transformation to Eq. 3.277 to get the
following expression

i4πce ∂f0 ρ̃ ω
(iv · k + p)f˜1 = fˆ1 (0) + k·
µωm ∂v p + ω 2
2
" #
4πe ∂f0 (pJ̃ − Ĵ(0)) ω
+ ·
ωm ∂v p2 + ω 2
" #
i4πce ∂f0 v · J̃ ω
− k·
ωm ∂v p2 + ω 2
" #
i4πce ∂f∂v · J̃ ω
0
+ v·k
ωm p2 + ω 2
(3.280)
where ρ̃ = ρ̃(k, v, p) and J̃ = J̃(k, v, p) are the corresponding expressions
in the Fourier-Laplace space. Rearranging terms on this expression, one
gets the solution
ifˆ1 (0) 4πce
f˜1 (k, v, p) = − + ×
(v · k − ip) µm(p2 + ω 2 )(v · k − ip)

∂f0 iµ ∂f0 ∂f0 ∂f0
× k· ρ̃ − · pJ̃ − Ĵ(0) − µ k · (v · J̃) + µ(v · k)
∂v c ∂v ∂v ∂v
(3.281)
The dispersion relation and other physical properties can be obtained by
substituting this expression into the definitions
Z Z
ρ̃ = e f˜1 (k, v, p)dv and J̃ = e vf˜1 (k, v, p)dv . (3.282)
<3 <3
3.12.2 Diluted linear, dispersed and nonlocal plasma

In this case, Maxwell’s equations are given by
1 ∂B 4π 1 ∂D
∇ · D = 4πρ , ∇ × E = − , ∇·B=0 , ∇×H= J+
c ∂t c c ∂t
(3.283)
where the relations between the fields E, B and D, H are given by
D = ? E and B = µ ? H, being and µ the permittivity and permeability
functions, and Rthe star product is just the space-time convolution,
(f ? g)(x, t) = f (x − x0 , t − t0 )g(x0 , t0 )dx0 dt0 ,
which characterizes the linear, dispersed and nonlocal behavior of the sys-
tem. We can use the Fourier-Laplace transformation on these equations,
Z Z ∞
1
f˜(k, p) = FL [f (x, t)] = eik·x−pt f (x, t)dxdt, (3.284)
(2π)3/2 <3 0
having the properties

i) FL [f1 + αf2 ] = FL [f1 ] + αFL [f2 ]
ii) FL [∂f /∂xl ] = ikl f˜(k, p)
iii) FL [∂f /∂t] = pf˜(k, p) − fˆ(k, 0)
iv) FL [f ? g] = f˜ · g̃ ,
with fˆ being just the Fourier transformation of f , and p is in general a
complex number. These Maxwell equation are transformed as
ik · E
e = 4π ρ̃/˜, ik × E e = − 1 (pB e −Bb 0) (3.285)
c
and
ik · B
e =0, ik × Be = 4π µ̃ J̃ + µ̃˜
e b
(pE − E0 ). (3.286)
c c
where we have defined E b 0 = Ê(k, 0) in the Fourier space, similarly with
B0 . From Eq. 3.272 and Eq. 3.273, one has that E
b b0 = B
b 0 = 0. Now, using
the same vector identity a × (b × c) = (a · c)b − (a · b)c we used before, it
follows that

−1 4π ρ̃ 4π µ̃p
E(k,
e p) = 2 i k + J̃ (3.287)
k 2 + µ̃˜2p ˜ c2
c
4π µ̃/c
B(k, p) = i 2 k × J̃, (3.288)
e
k + µ̃˜c2p
2
where ρ̃ and J̃ are, of course, given in terms of the transformed distribution

function f˜ as
Z Z
ρ̃(k, p) = ρi − e f˜(k, v, p)dv , J̃(k, p) = Ji − e vf˜(k, v, p)dv
<3 <3
(3.289)
At first order approximation, we can calculate the distribution function by
taking the Fourier-Laplace transformation to Eq. 3.265, which brigs about
the expression
fˆ1 (k, v, 0) 1 ∂f0
f˜1 (k, v, p) = − F(k,
e v, p, f˜1 ) · , (3.290)
iv · k + p iv · k + p ∂v
where F
e is given by

˜ e e 1 e
F(k, v, p, f1 ) =
e E+ ×B (3.291)
m c
where Ee and B e are given by Eq. 3.287 and Eq. 3.288. By substituting
one expresion into another would appears a relation which brings about the
dispersion relation of the system.
3.12.3 Non Dilute Plasma

In this case there are interactions among the different species of ions and
electrons which the most probable is the two bodies interaction (ion-ion,
ion-electron, or electron-electron). However, it is assume that the probabil-
ity of having a particular species in a element of phase-space-time dxdvdt
is independent for each specie. In this way, the distribution function of the
species α, fα (x, v, t), having a density
Z
ρα (x, t) = fα (x, v, t)dv , (3.292)
<3
satisfies the so called Boltzmann’s equation (under interaction with elec-
tromagnetic field)

∂fα e h v i ∂f
α ∂f ∂fα
v· + E+ ·B · + = , (3.293)
∂x m c ∂v ∂tα ∂t c
where the term on the right hand side denotes the rate of change of fα
due to collisions. Due to the effect of the screening (Debye length), the
Coulomb interaction last for a time τ ∼ λD /v (being v the relative veloc-
ity of ions in collision), and it is assumed that during this time the density
distributions of the ions in collision do not change significantly, with a max-
imum parameter of impact given by the Debye length. There are several
approximations for this collision term (see reference J. Sheffield and ref-
erence S. Chandrasekhar, and reference C. Cercignani ), and we limited
ourselves to the case when this term depend only of the same distribution
function fα (for example in the Krook model (∂fα /∂t)c = −ναβ (fα − fα0 )
with ναβ being the collision frequency between the species α and β, and
fα0 is the relaxed distribution function). Therefore, dropping the index α
in Eq. 3.294, the Boltzmann’s equation looks as
∂f e h v i ∂f ∂f
v· + E+ ·B · + = C(f ) , (3.294)
∂x m c ∂v ∂t
Consider an external electromagnetic field given by the following expres-
sions

E = Ex (t), Ey (t), Ez (t) , and B = 0, 0, B , (3.295)
where B is constant. Then, substituting these fields in Eq. 3.294, one gets
the following equation

∂f ∂f ∂f ∂f e vy B ∂f
+ vx + vy + vz + Ex +
∂t ∂x ∂y ∂z m c ∂vx

e vx B eEz ∂f
+ Ey − + = C(f ) . (3.296)
m c m ∂vz
This equation represents a quasi-linear PDEFO defined in <7 , and the

equations for its characteristics are given by (in parametric form)

dx dy dz dvx e vy B
= vx , = vy , = vz , = Ex +
dt dt dt dt m c

dvy e vx B dvz e df
= Ey − , = Ez , = C(f ) (3.297)
dt m c dt m dt
Differentiating the fourth and fifth equations and using them again, one
gets the decoupled equations
d2 vx d 2 vy

2 e 2 e
+ ω vx = Ėx + Ey , and + ω vy = Ėy − Ex ,
dt2 m dt2 m
(3.298)
where ξ˙ means dξ/dt. Since the right hand side of these equation are
functions depending on time, the solutions of these equation are given by
1 t
Z
vx = c1 cos ωt + c2 sin ωt + g1 (s) sin ω(t − s) ds (3.299)
ω
and
Z t
1
vy = c3 cos ωt + c4 sin ωt + g2 (s) sin ω(t − s) ds , (3.300)
ω
where the functions g1 and g2 have been defined as

e e
g1 (t) = Ėx + Ey and g2 (t) = Ėy − Ex . (3.301)
m m
Thus, the first two equations of 3.297 are readily integrated as

Z Z t
c1 c2 1
x= sin ωt − cos ωt + dt g1 (s) sin ω(t − s) ds (3.302)
ω ω ω
and
Z Z t
c3 c4 1
y= sin ωt − cos ωt + dt g1 (s) sin ω(t − s) ds (3.303)
ω ω ω
The integration over the z-dependence is straightforward and is given by
e t
Z Z Z t
e
vz = Ez (s) ds + c5 and z = dt Ez (s) ds + c5 t + c6 .
m m
(3.304)
3.13. Decoherence in a quantum system 109
Therefore, the characteristics are given by

Z t Z Z t
c1 = vx + xω + cos ωt g1 (s) sin ω(t − s)ds + sin ωt dt g1 (s) sin ω(t − s)ds ,
Z t Z Z t
c2 = vx − xω + sin ωt g1 (s) sin ω(t − s)ds − cos ωt dt g1 (s) sin ω(t − s)ds ,
Z t Z Z t
c3 = vy + yω + cos ωt g2 (s) sin ω(t − s)ds + sin ωt dt g2 (s) sin ω(t − s)ds ,
Z t Z Z t
c4 = vy − yω + sin ωt g2 (s) sin ω(t − s)ds − cos ωt dt g2 (s) sin ω(t − s)ds ,
e t
Z
c5 = vz − Ez (s)ds ,
m
Z t
et t
Z Z
e
c6 = z − dt Ez (s)ds − vz t + Ez (s)ds .
m m
(3.305)
Finally, the integration of the last equation in 3.297 can be expressed as
Z
df
= t + g c(x, v, t) , (3.306)
C(f )
where the vector c has been defined as c = (c1 , c2 , c3 , c4 , c5 , c6 ), and g is an
arbitrary function. For the Krook model we can write the general solution
as
f (x, v, t) = fα0 + G c(x, v, t) e−ναβ t ,

(3.307)
where G is an arbitrary function which is determined by initial conditions.
3.13 Decoherence in a quantum system
In the Schrödinger picture of the quantum mechanics one can solve a prob-
lem either with the Schrödinger equation for the wave function |Ψi,
∂|Ψi
i~ = H|Ψi
b , (3.308)
∂t
where |Ψi is a ket vector element of a finite or infinite Hilbert space (see
reference A. Messiah and reference P.A.M. Dirac), and H b is an Hermitian
operator associated to the Hamiltonian of the system, or we can solve a
problem with the von Neuman equation for the density matrix operator ρ,
∂ρ
i~ = [H,
b ρ] , (3.309)
∂t
where ρ = |ΨihΨ| is an Hermitian operator with ρ ≥ 0, tr(ρ) = 1, and

tr(ρ2 ) = 1 (for pure state, ρ2 = ρ, which is solution of 3.308), [a, b] = ab−ba
is the commutator of two operators. If the relation ρ2 6= ρ is satisfied,
one says the density matrix corresponds to a mixed state. The evolu-
tion of the system is unitary in both cases, that is, there is an unitary
operator U (U † U = I) such that teh evolution of the systems is given
|Ψ(t)i = U (t)|Ψ0 i or ρ(t) = U (t)ρ0 U † (t). Although must of the quantum
system (QS) are considered insulated from environment, that is not gener-
ally true in practice, and the interaction of the quantum system with the
environment brings about the relaxation and decoherence of the system.
One says that within a quantum system has appeared decoherence when
the non diagonal elements of the density matrix become zero (the matrix
becomes diagonal). The study of a QS with his environment is called open
quantum system (OQS), and the Hamiltonian of a OQS would be of the
form H b = H bs + Hbe + H b se , where H
b se contains the interaction between
system and environment. When this Hamiltonian is introduced in 3.308
and the trace over the environment variables is taken, the resulting density
matrix is called ”reduced density matrix,” , and its evolution is not any
longer unitary. The resulting evolution equation is called ”master equation
of motion,” and it used to be of Limdblad type of equation (see reference
G. Limdblad, and see reference H.P. Breuer and F. Petruccione),
∂ρ b s , ρ] + L(ρ) ,
i~ = [H (3.310)
∂t
where ρ is the reduced density matrix, and L is a linear operator which
contains the terms resulting from the environment and system-environment
interaction at some approximation . For a 1-D, QS with continuos variables
interacting with the environment at room temperature in a Markovian pro-
cess (without memory), the master equation is written as (see A.O. Caldeira
and A.J. Leggett),
∂ρ b s , ρ]+ β {p, z}, ρ − i Dβ[z, [z, p]]+2β([z, ρp]−[p, ρz]) , (3.311)

i~ = [H
∂t 2 ~
where {a, b} = ab + ba is the anti-commutator of two operators, z and p
are the position and momentum of the particle, β represents the dissipa-
tion coefficient, and D = 2kB T /~ represents the diffusion coefficient which
depends on the temperature T .
For a Josephson junction device coupled to LC oscillator via a capaci-

tor, Cm which is coupled to a transmission line via another capacitor, Cc ,
and with a oscillator capacitance Co = C + Cm , a Hamilton for this quan-

tum system (ignoring the transmission line charge) can be written as (see
reference G. Johanson et al, see also reference Y. Makchlin et al)
b S = 1 p2 + z 2 − σz − ηp2 σz ,

H (3.312)
2 2
where p is proportional to the charge of the magnetic flux oscillator,
1/4
p = qc L/Co ~2 , the variable z is proportional to this magnetic flux,
2 1/4

z = φc Co /L~ , which are both conjugated, [z, p] = i. The evolution
√
is given in terms of the LC value of the oscillator, τ = t/ LCo , and the
√
Hamiltonian has been written in a normalized way, H b S = LCo H/~. b The
parameter η = gc Co /2 measures the coupling between the oscillator and the
two states of the superconductor junction (current flowing in one direction
or the its opposed direction, like the two states spin one half particle),
√
and = EJ LCo /~ is the normalized energy in the junction (qubit). The
reduced density matrix will depend on two continuous variables, z and z 0 ,
and two discrete variables, s and s0 , ρss0 (z, z 0 ) = hz, s|ρ|z 0 , s0 i. Using 3.312
in 3.311, the resulting master equation (after some rearrangements) is given
by

∂ρss0 i i
= (∂zz − ∂z0 z0 ) − (z 2 − z 02 ) − iη(s∂zz − s0 ∂z0 z0 )
∂τ 2 2

β 0 0 2
− (z − z )(∂z − ∂z0 ) − Dβ(z − z ) ρss0 , (3.313)
2
where s, s0 = ±1/2. This equation represents a partial differential equation
of second order. However, let us make first the following change of variables
(see reference G. López and P. López)
r = z − z0 , R = (z + z 0 )/2, (3.314)
getting the following expression for this equation as
s − s0

∂ρss0
= i∂rR − irR − βr∂r − iη (s − s0 )∂rr + (s + s0 )∂rR +

∂RR
∂τ 4

−βr∂r − Dβr2 ρss0 . (3.315)
Now, taking Rthe Fourier transformation with respect the variable R,

ρ̂ss0 (r, k, τ ) = < eikR ρss0 (r, R, τ )dR, the following equation is gotten
∂ ρ̂ss0 ∂ ρ̂ss0 ∂ ρ̂ss0 ∂ 2 ρ̂ss0
= k − βr − kη(s + s0 ) − iη(s − s0 )

−r
∂τ ∂r ∂k ∂r2
0
s − s
− Dβr2 + k 2 ρ̂ss0 .

(3.316)
4
With respect the current flowing in the superconducting joint (character-

ized by the discrete variables s and s0 ), this equation is divided in the
diagonal matrix elements (s = s0 ), which satisfies the equation
∂ ρ̂s ∂ ρ̂s ∂ ρ̂s

+ βr − k + 2kηs −r = −Dβr2 ρ̂s , (3.317)
∂τ ∂r ∂k
and the non diagonal matrix elements (s = −s0 ), which satisfies the equa-
tion
∂ ρ̂0s ∂ ρ̂0s ∂ ρ̂0 sk 2 0 ∂ ρ̂0

− r s = − Dβr2 + ρ̂s − i2sη 2s .

+ βr − k (3.318)
∂τ ∂r ∂k 2 ∂r
Diagonal elements: Let us deal with Eq. 3.317 which represents a linear
PDEFO defined in <3 . The equations for its characteristics in parametric
form are given by
dk dr dρs
=r, = βr − k + 2kηs , = −Dβr2 ρs . (3.319)
dτ dτ dτ
Making the differentiation of the first term and using the second one, one
gets
d2 k dk
+ (1 − 2ηs)k − β =0 (3.320)
dτ 2 dτ
k(τ ) = eβτ /2 a1 sin ωτ + a2 cos ωτ ,

(3.321)
where ω has been defined as
p
ω= 1 − β 2 /4 − 2ηs . (3.322)
Therefore, using the first term of 3.319, we also have
β β
r(τ ) = eβτ /2 a1 (ω cos ωτ + sin ωτ ) + a2 (−ω sin ωτ + cos ωτ ) . (3.323)

2 2
From these expressions we get the characteristics
e−βτ /2 β
a1 = k(ω sin ωτ − cos ωτ ) + r cos ωτ (3.324)
ω 2
e−βτ /2 β
a2 = k(ω cos ωτ + sin ωτ ) − r cos ωτ (3.325)
ω 2
Using Eq. 3.323 in the last equation of 3.319, making the integration and
the above characteristics, it follows that
2 2 2 2
ρ̂s (r, k.τ ) = G a1 (τ ), a2 (τ ) e−D[(β + 4ω )k + 4r ]/8 ,

(3.326)
where G is an arbitrary function depending on the characteristics and is

determined by the initial conditions. Assume that ρ̂s (r, k, 0) = ρ̂0 (r, k),
then, since a1 (0) = (r − βk/2)/ω and a2 (0) = k, one gets
2 2 2 2
ρ̂0 (r, k) = G (r − βk/2)/ω, k e−D[(β + 4ω )k + 4r ]/8 .

Now, defining the parameters ξ1 = (r − βk/2)/ω and ξ2 = k, we obtain the

functionality of G as
2 2 2 2
G(ξ1 , ξ2 ) = ρ̂0 ξ1 ω + βξ2 /2, ξ2 e−D[(β + 4ω )ξ2 + 4(ξ1 ω + βξ2 /2) ]/8 .

Thus, the solution of our problem is given by

ρ̂s (r, k, τ ) = ρ̂0 a1 (τ )ω + βa2 (τ )/2, a2 (τ ) ×
2 2 2 2
e−D[(β + 4ω − p1 (τ ))k + (4 − p2 (τ ))r + p3 (τ )kr]/8 , (3.327)
where one has p1 (0) = β 2 + 4ω 2 , p2 (0) = 4, and p3 (0) = 0, and these

functions have been defined as
e−βτ (β 2 + 4ω 2 ) 2
β + 4ω 2 − β 2 cos 2ωτ + 2βω sin 2ωτ (3.328)

p1 (τ ) = 2
4ω
e−βτ
4(β 2 + 4ω 2 ) − 4β 2 cos 2ωτ − 8βω sin 2ωτ

p2 (τ ) = (3.329)
4ω 2
e−βτ
−4β(β 2 + 4ω 2 ) + 4β(β 2 + 4ω 2 ) cos 2ωτ

p3 (τ ) = 2
(3.330)
4ω
Once we have gotten the solution in the Fourier space, the solution in the
normal space (R) is gotten by using the inverse Fourier transformation,
√ Z
ρs (r, R, τ ) = F −1 [ρs (r, k, τ )] = 2π e−ikR ρ̂s (r, k, τ )dk.
<
However, one can calculate some dynamical variables without doing this,
for example
Z

hzi(τ ) = tr(ρs z) = z ρ1/2 (z, z, τ ) + ρ−1/2 (z, z, τ ) dz. (3.331)
<
To see this, note that ρs (z, z, τ ) implies from 3.314 that r = 0 and R = z,
that is, it follows that
Z

hzi(τ ) = R ρ1/2 (0, R, τ ) + ρ−1/2 (0, R, τ ) dR
Z<
RF −1 ρ̂1/2 (0, k, τ ) + ρ−1/2 (0, k, τ ) dR

=
<
√ Z Z
e−ikR ρ̂1/2 (0, k, τ ) + ρ−1/2 (0, k, τ ) dk

= 2π R dR
< <
√ Z

∂ ρ̂1/2 (0, k, τ ) ∂ρ−1/2 (0, k, τ )
Z
= −i 2π dR e−ikR + dk
< < ∂k ∂k
√

∂ ρ̂1/2 (0, k, τ ) ∂ρ−1/2 (0, k, τ )
= −i 2π +
∂k ∂k k=0
Therefore, the expectation value of the magnetic flux can be calculated by

knowing the reduced density matrix in the Fourier space as
" #
√ ˆ −1/2 (0, k, τ )
∂ ρ̂1/2 (0, k, τ ) ∂ρ
hzi(τ ) = −i 2π + . (3.332)
∂k ∂k
k=0
Assume that the initial reduced density matrix is given by

(0) 1 11 1 0 2 0 2
ρss0 (z, z 0 ) = √ eipo (z−z )−(z−zo ) /2−(z −zo ) /2 . (3.333)
2 11 2π
Then, the diagonal element in the coordinates (r, R) is
1 2 2
ρ0 (r, R) = √ eipo r−(R+r/2−z−o) /2−(R−r/2−zo ) /2 , (3.334)
2 2π
and its Fourier transformation is given by
1 2 2
ρ̂0 (r, k) = √ eipo r−r /4+ikzo −k /4 . (3.335)
2 2π
Writing its dependence in terms of a1 and a2 , it follows that
1 2 2
ρ̂0 (a1 ω + βa2 /2, a2 ) = √ eipo [a1 ω+βa2 /2]−[a1 ω+βa2 /2] /4+ia2 zo −a2 /4 .
2 2π
(3.336)
since from Eq. 3.332 this expression now must be valuated at r = 0, we
see from 3.324 and 3.325 that a1 (r = 0) = χ1 (τ )k and a2 (r = 0) = χ2 (τ )k,
where
e−βτ /2 β
χ1 (τ ) = (ω sin ωτ − cos ωτ ) (3.337)
ω 2
e−βτ /2 β
χ2 (τ ) = (ω cos ωτ + sin ωτ ) (3.338)
ω 2
In this way, one gets

1 2 2
ρ̂0 (a1 ω +βa2 /2, a2 )r=0 = √ ei{po [χ1 ω+βχ2 /2]+zo }k−{[χ1 ω+βχ2 /2] +1}k /4 .

2 2π
(3.339)
Using this expression in Eq. 3.327, and after valuating at r = 0 and making
its differentiation and valuating at k = 0, we get

po
hzi(τ ) = zo + ωχ1 + βχ2 /2 s=1/2 + ωχ1 + βχ2 /2 s=−1/2 . (3.340)
2
Note from 3.322 that ω also depends on s.
Non diagonal elements: Let us write the non diagonal elements of the
reduced density matrix as ρ̂0s = fˆ + iĝ, and let us separate the real and
imaginary parts from the Eq. 3.318. The equations for f and g can be
written as
∂ 2 ĝ ∂ 2 fˆ
L(fˆ) = 2ηs 2 , L(ĝ) = −2ηs 2 , (3.341)
∂r ∂r
where L is the linear operator defined as
∂ ∂ ∂ sk 2
L= + (βr − k) +r + Dβr2 + . (3.342)
∂τ ∂r ∂k 2
Eq. 3.341 can be solved by iteration method in any compact set of <3 ,
related to the variables r, k and τ . Assuming one has this compact set, the
equations to be solved are
∂ 2 ĝn−1 ∂ 2 fˆn−1
L(fˆn ) = 2ηs 2
, L(ĝn ) = −2ηs , n ≥ 1, (3.343)
∂r ∂r2
and such that L(fˆ0 ) = L(ĝ0 ) = 0. The equation defined by L(u) = 0 is
given by
∂u ∂u ∂u sk 2
+ (βr − k) +r = − Dβr2 + u (3.344)
∂τ ∂r ∂k 2
and represents a linear PDEFO defined in <3 , u = u(r, k, τ ). The equations
for its characteristics in parametric form are given by
dk dr du sk 2
= r, = βr − k, and = − Dβr2 + u. (3.345)
dτ dτ dτ 2
Making the differentiation with respect to τ of the first two equations above,
one gets
d2 k dk
2
+k−β =0 (3.346)
dτ dτ

k(τ ) = eβτ /2 b1 sin ωo τ + b2 cos ωo τ ,

(3.347)
where ωo has been defined as
p
ωo = 1 − β 2 /4. (3.348)
In this way, the solution for the variable r is

βτ /2 β β
r(τ ) = e sin ωo τ + ωo cos ωo τ b1 + cos ωo τ − ωo sin ωo τ b2 .
2 2
(3.349)
From these expressions we get the characteristics
e−βτ /2 β
b1 = k(ωo sin ωo τ − cos ωo τ ) + r cos ωo τ (3.350)
ωo 2
e−βτ /2 β
b2 = k(ωo cos ωo τ + sin ωo τ ) − r cos ωo τ (3.351)
ωo 2
making use of these expressions, the integration of the last term in 3.345
can be done, bringing about the solution
2
+α2 r 2 −2α3 kr]
u(r, k, τ ) = G(b1 , b2 )e−[α1 k , (3.352)
where G is an arbitrary function depending on the characteristics, and αi
for i = 1, 2, 3 has been defined as
D 2 s(5β 2 + 4ωo2 )
α1 = (β + 4ωo2 ) + (3.353)
8 4β(β 2 + 4ωo2 )
D s
α2 = + 2 (3.354)
2 β + 4ωo2
s
α3 = 2 . (3.355)
β + 4ωo2
The functionality of G is determined by the initial conditions, u(r, k, 0) =
uo (r, k), and proceeding in the same way as we did for the diagonal case,
one gets
2 2
−2α3 ξ2 (ωo ξ1 +βξ2 /2)
G(ξ1 , ξ2 ) = uo (ωo ξ1 + βξ2 /2, ξ2 )eα1 ξ2 +α2 (ωo ξ1 +βξ2 /2) ,
where ξ1 and ξ2 are any variables. Therefore, the solution of Eq. 3.344 can
be written (after some arrangements) as

u r, k, τ = uo ωo b1 (τ ) + βb2 (τ )/2, b2 (τ ) ×
2 2
e{(α1 − q1 )k + (α2 − q2 )r − 2(α3 + q3 )kr} , (3.356)
where the functions qi for i = 1, 2, 3 have been defined as

e−βτ

4β 2 α1 + β 4 α2 − 4β 3 α3 + 8β 2 ωo2 α2 − 16βωo3 α3 + 16ω 2 α2 sin2 ωo τ

q1 (τ ) =
16ωo

2 2 2 3

+16ωo α1 cos ωo τ + 8βωo α1 − 4β ωo α3 − 16ωo α3 sin 2ωo τ (3.357)
e−βτ

16ωo2 α2 cos2 ωo τ + 16α1 + 4β 2 α2 − 16βα3 sin2 ωo τ

q2 (τ ) = 2
16ωo

+ 16ωo α3 − 8βωo α2 sin 2ωo τ (3.358)
e−βτ

−16ωo2 α3 cos2 ωo τ + −2β 2 α2 + 8β 2 α3 + 16ωo2 α3 − 8β(α1 + α2 ωo2 ) sin2 ωo τ

q3 (τ ) = 2
16ωo

−2ωo −β 2 α2 + 4(α1 − α2 ωo2 sin 2ωo τ .

(3.359)
These functions have the initial values
q1 (0) = α1 , q2 (0) = α2 , and q3 (0) = −α3 . (3.360)
Having solved the problem L(u) = 0, we can now proceed to solve Eq. 3.343
for any n ≥ 1. Let us solve first the case for n = 1, and the generalization
will be given straightforwardly. The equation defines by 3.343 is written as
∂ fˆ1 ∂ fˆ1 ∂ fˆ1 sk 2 ˆ ∂u

+ (βr − k) +r = − Dβr2 + f1 + 2ηs 2 . (3.361)
∂τ ∂r ∂k 2 ∂r
This expression represents a linear PDEFO defined in <3 , where the equa-
tions for its characteristics are given by Eq. 3.345 but the last one which
is changed by
dfˆ1 sk 2 ˆ ∂2u
= − Dβr2 + f1 + 2ηs 2 . (3.362)
dτ 2 ∂r
The solution for r and k is the same as Eq. 3.347 and Eq. 3.349, with the
same angular frequency 3.348 and the same characteristics 3.350 and 3.351.
The solution of Eq. 3.362 is readily given by
e 1 , b2 )e−[α1 k2 +α2 r2 −2α3 kr]
fˆ1 (r, k, τ ) = G(b
Z τ 2
2 2 ∂ u −[α1 k2 (σ)+α2 r2 (σ)−2α3 k(σ)r(σ)]
+2ηse−[α1 k +α2 r −2α3 kr] e dσ
0 ∂r2
(3.363)
The arbitrary function G e is determined as before. Thus, the solution of Eq.

3.361 is gotten as
2 2
fˆ1 (r, k, τ ) = uo ωo b1 + βb2 /2, b2 e−[(α1 −q1 )k +(α2 −q2 )r −2(α3 +q3 )kr]

Z τ 2
2 2 ∂ u −[α1 k2 (σ)+α2 r2 (σ)−2α3 k(σ)r(σ)]
+2ηse−[α1 k +α2 r −2α3 kr] e dσ
0 ∂r2
(3.364)
where the functions qi = qi (τ ) for i = 1, 2, 3 are given as above. A similar
solution (changing η to −η) is obtained for the imaginary part ĝ1 of Eq.
(3.341). In addition, It is straightforward to see the general solution of Eq.
3.343 which can be written as
2 2
fˆn (r, k, τ ) = uo ωo b1 + βb2 /2, b2 e−[(α1 −q1 )k +(α2 −q2 )r −2(α3 +q3 )kr]

Z τ 2
2 2 ∂ ĝn−1 2 2
+2ηse−[α1 k +α2 r −2α3 kr] 2
e−[α1 k (σ)+α2 r (σ)−2α3 k(σ)r(σ)] dσ
0 ∂r
(3.365)
and
2 2
ĝn (r, k, τ ) = ũo ωo b1 + βb2 /2, b2 e−[(α1 −q1 )k +(α2 −q2 )r −2(α3 +q3 )kr]

!
∂ 2 fˆn−1
Z τ
−[α1 k2 +α2 r 2 −2α3 kr] 2 2
−2ηse 2
e−[α1 k (σ)+α2 r (σ)−2α3 k(σ)r(σ)] dσ
0 ∂r
(3.366)
These solutions are given in the Fourier space, and one still need to make
the inverse transformation to get the solution in our variables (r, R, τ ).
Since the solution for any ”n” depends on the first solution f1 and g1 , the
decoherence behavior (ρ0s → 0 for s = 1/2, −1/2) depends mainly on the
decoherence of these first elements (f1 → 0 and g1 → 0).
3.14. Problems 119
3.14 Problems
3.1 Write down the PDEFO associated to the constant of motion, K, of

the dynamical system dx/dt = v and dv/dt = (−g + αv 2 /m), where α is
a non negative constant, and m represents the mass of a particle. i) Show
that the characteristic can be written as
αv 2

mg
C=− ln 1 − + gx
2α mg
ii) Show that the solution of the equation for the constant of motion is
given as K = G(C), where G is a general function.
iii) Show that by choosing G(C) = mC and G(C) = −(mg/2α)e−2αC/mg −
m2 g/2α, one gets the following two equivalents constant of motions
m2 g αv 2

K1 (x, v) = − ln 1 − + mgx
2α mg
and
m2 αv 2 m2 g

K2 (x, v) = −g + e−2αx/m + .
2α m 2α
iv) Show that the following limit is gotten limα→0 Ki = mv 2 /2 + mgx for
i = 1, 2.
v) Show that we can get two Lagrangian for the system given by
m2 g αv 2
r r
mg α
L1 (x, v) = m v arctanh v + ln 1 − − mgx
α mg 2α mg
and
m2 αv 2 m2 g

L2 (x, v) = g+ e−2αx/m − .
2α m 2α
vi) Show that limα→0 Li = mv 2 /2 − mgx for i = 1, 2.
vii) Show that, given these two Lagrangian, one gets the following two
Hamiltonian for the system
m2 g
r
2 p α
H( x, p) = − ln 1 − tanh + mgx
2α m mg
and
m2 αp2 4αx/m −2αx/m m2 g

H2 (x, p) = −g + e e + .
2α m3 2α
viii) Show that limα→0 Hi = p2 /2m + mgx for i = 1, 2.
3.2 Find the solution of the PDEFO found in exercise (12) of chapter 1,
assuming that the equation
dx dv
=
v F (x, v)
brings about the characteristic curve C = C(x, v), and show that the solu-
tion for L(x, v) (J = Lvv ) can be written as
Z v Z v0 R v00 Fṽ
0
L(x, v) = dv dv 00 A C(x, v 00 ) e− dṽ/F (x,ṽ)
+ B1 v + B2 ,
where B1 and B2 are constants, and A is an arbitrary function of the

characteristic C.
3.3 The equation of motion of a one-degree of freedom relativistic particle

of mass m in a dissipative medium characterized by a quadratic dissipation
force with dissipation constant α, −α(dx/dt)2 and v ≥ 0, can be written in
Newton form as
d2 x
m = −αv 2 (1 − v 2 /c2 )3/2 , c is the speed of light .
dt2
i) Find the constant of motion associated to this system by solving its as-
sociated PDEFO.
ii) Determine this constant of motion such that limc→∞ K = mv 2 /2.
iii) If it is possible, find the Lagrangian of the system by using Eq. (3.17a).
If it is not possible, determine the main difficulty.
iv) If it is possible, find the Hamiltonian of the system by using Eq. (3.29).
If it is not possible, determine the main difficulty.
v) Since v/c ≤ 1, write down the constant of motion in Taylor series expan-
sion around v/c = 0, and using this expression, repeat the steps (iii) and
(iv).
3.4 Show that the constant of motion associated to the dynamical system
dx dv α
=v , = −ω 2 x − v
dt dt m
can be given by the following expression
m 2
v + 2ωα xv + ω 2 x2 e−2ωα Gα (v/x,ω) ,

K(x, v) =
2
3.14. Problems 121
where α is the dissipative constant of the system, ωα is defined as ωα =

α/2m, and Gα is defined as
 √ 2 2
1 ωα +v/x− ωα −ω


 √ 2 −ω 2
ln √ 2 −ω 2
, if ω 2 < ωα2

 2 ωα ωα +v/x+ ωα






1
Gα = ω2 +v/x , if ω 2 = ωα2







1 ωα +v/x
√ √

arctan , if ω 2 > ωα2


 ω2 −ω2 ω 2 −ω 2
α α
corresponding to the weak, critical, and strong dissipation cases. Show the
difficulty to find the Lagrangian and the Hamiltonian of the system, even
at first order in the parameter ωα (very weak dissipation).
3.5 Find the eigenvalues of the operator L+ ( Eq. (3.41a)), and look for
a polynomial solution.
3.15 References
3.1 G. López, Constant of Motion, Hamiltonian, and Lagrangian for Au-

tonomous Systems in Hyperbolic Flat Spaces, SSCL-Preprint-150, August
1993.
3.2 A.R. Edmonds, Angular Moment in Quantum Mechanics, Princeton

University Press, 1957. Chap. 2.
3.3 B.L. Vander Waerden, The Group Method in Quantum Mechanics,

Springer-Verlag, 1938. Chap. 3.
3.4 G. López, Analytical Approximation to the Turn-to-Turn Quench

Propagation, SSCL-309, September 1990.
3.5 K. Huang, Statistical Mechanics, John Wiley & Sons, 1987. Chap.6-7.
3.6 P. Srivastava, TCP (truncated compound Poisson) Process for Mul-

tiplicity Distribution in High Energy Collisions, International Center for
Theoretical Physics, Ic/88/127.
3.7 E.D. Courant and H.S. Snyder, Ann. Phys., 3 (1958) 1.
3.8 M. Sands, The Physics of Electron Storage Rings, and Introduction,

SLAC-121, Nov. 1970.
3.9 G. López and S. Chen, Tune Shift Effect Due to Multipole Longitudi-
nal Periodic Structure in the Superconducting Dipole Magnets, SSCL-550,
October 1991.
3.10 H. Goldstein,Classical Mechanics, Addison-Wesley, Reading MA,

1950.
3.15. References 123
3.11 A.W. Chao, Physics of Collective Beam Instabilities in High Energy

Accelerators, John Wiley and Sons, Inc., 1993.
3.12 P.J. Bryan and K. Johnsen,The Principles of Circular Accelerators

and Storage Rings, Cambridge University Press, 1993.
3.13 A.W. Chao, Coherent Instabilities of a Relativistic Bunched Beam,

SLAC-PUB-2946, June 1982.
3.14 G. López, Ann. Phys., 251, (1996) 1.
3.14 N.M. Atakishiyev, T.H. Seligman, K.B. Wolf, Proceedings of the IV

Wigner Symposium, World Scientific, 1995. Page 250.
3.15 A.N. Kolmogorov, Dokl. Akad. Nauk. SSSR, 98, (1954) 527.
3.16 V.I. Arnold, Dokl. Akad. Nauk. SSSR, 156, (1964) 9.
3.17 L. Landau, On the vibration of the electronic plasma. J. Phys. USSR

10,25, (1946). (JETP 16, 574, (1946)).
3.18 V. Maslov and M. Fedoryuk, The linear theory of Landau damping,

Mat. Sb. 127, 445 (1985).
3.19 J.R. Reitz, Foundation of Electromagnetic Theory, Addison Wesley,

(1969), chapter 14.
3.20 J. Scheffield, Plasma Scattering of Electromagnetic Radiation, Aca-

demic Press (1975), appendix 2.
3.21 S. Chandrasekhar,Plasma Physics, The University of Chicago Press

(1960), chapter VII.
3.22 J.D. Jackson,Classical Electrodynamics (1999), chapter 14.
3.23 C. Cercignani, The Boltzmann Equation and Its Applications,

Springer-Verlag (1988).
3.24 A.O. Benz, Plasma Astrophysics, Kinetic Processes in Solar and

Stellar Coronae, Kluwer Academic Publishers, (2002), chapter 5.
3.25 A. Messiah, Quantum Mechanics, John Wiley & Sons, (1976), vol.
I, II.
3.26 P.A.M. Dirac, The Principles of Quantum Mechanics, Oxford Uni-

versity Press, (1976).
3.27 G. Lindblad, On the generators of quantum dynamical semigroups,

Commun. Math. Phys. 48, (1976) 119.
3.28 A.O. Caldeira and A.J. Legget, Path integral approach to quantum
Brownian motion, Physica A, 121 (1983) 587.
3.29 G. Johanson, L. Tornberg, and C.M. Wilson, Phys. Rev. B., 74,
(2006) 100504(R).
3.30 H.P. Breuer and F. Petruccione, The Theory of Open Quantum

Sysytems, Oxford University Press, (2009).
3.31 Y. Makhlin, G. Schön, and A. Shnirman, Rev. Mod. Phys., 73,

(2001) 357.
3.15. References 125
3.32 G. López and P. López, internal report, University of Guadalajara,

México,RI-2009-COC1, (2009).

Chapter 4
Nonlinear Partial Differential

Equations of First Order
In this chapter we will developed the theory of the nonlinear PDEFO de-
fined on <2 and on <n .
4.1 Non-linear PDEFO for Functions Defined in R2
The PDEFO for functions defined in the plane (x, y) have the general form
F (x, y, z, q, p) = 0, (4.1)
where z = z(x, y), and p and q are defined as
∂z ∂z
p= , q= . (4.2)
∂x ∂y
Let z = f (x, y) be the integral surface of Eq. (4.1), then at any point
P = (x, y, z) on this surface, Eq. (4.1) gives us a nonlinear relation between
p and q
φ(p, q) = 0 (4.3)
in such a way that there can exist more than one possible solution for p,
with q given (or the other way around), then the normal vector,
1
n̂ = (p, q, −1), (4.4)
(p2 + q 2 + 1)1/2
to the surface at this point is not unique defined (see Fig. 4.1). In this
way we obtain a one parameter family of normal-vectors to the surfaces at
the point P . Its envelope will be called the Monges’s cone of the normal
directions. As a consequence, a one parameter family of tangent planes
passing through the point P (x, y, z) is formed, Z − z = p(X − x) + q(Y − y),
and this envelope is called Monge’s cone of the tangent planes.
127
128 Nonlinear Partial Differential Equations of First Order
Fig. 4.1 Family of normal vectors generated by (4.4), the envelope is generated by the
non linear relation between p and q (4.2).
Let establish this point with particular examples. In the quasi-linear

case, see Eq. (2.15), we have
F (x, y, z, q, p) = a(x, y, z)p + b(x, y, z)q − c(x, y, z) = 0,
evaluating the coefficients a, b and c at a given point P , they become real

numbers and the above differential equation becomes a linear algebraic
equation between p and q. Then, knowing q at the point P (x, y, z), there
exist one and only one value for p, and this defines a unique normal direc-
tion. However, consider the non-linear partial differential equation
F (x, y, z, p, q) = an pn + an−1 pn−1 + ... + a1 p + bq − c = 0
where the coefficients ai i = 1, ..., n, b and c are arbitrary functions at x, y

and z. Evaluating the above equation at the point P (x, y, z), becomes an
algebraic equation of order n in p and knowing q there exists n-complex
possible values for p, that is, the normal of the surface at the point P (x, y, z)
is not uniquely defined.
Assume that we have the integral surface of Eq. (4.1) given by
Φ(x, y, z, a, b) = 0, (4.5)
where a and b are constants. The relation (4.5) defines a two parameters
family of surfaces. The envelope of this family will be also an integral
surface because the normal vector to the envelope surface coincides with the
4.1. Non-linear PDEFO for Functions Defined in R2 129
normal vector to one of the integral surface of the family. The determination
of this envelope is given by
∂Φ ∂Φ
Φ(x, y, z, a, b) = 0, = 0, = 0. (4.6)
∂a ∂b
We can obtain a one parameter of integral surface if we make b = b(a) and
in this case the envelope will be determined by
∂
Φ(x, y, z, a, b(a)) = 0, Φ(x, y, z, a, b(a)) = 0 (4.7a)
∂a
or
∂Φ ∂Φ ∂b
Φ(x, y, z, a, b(a)) = 0, + = 0. (4.7b)
∂a ∂b ∂a
Because these last relation contain an arbitrary parameter, this allows us
to obtain an integral surface which satisfies the Cauchy initial conditions.
Example 4.1. Find the envelope to the family of surfaces given by the
equation
(x − a)2 + y 2 − z 2 = b2 .
The family of surfaces (4.5) is
Φ(x, y, z, a, b) = (x − a)2 + y 2 − z 2 − b2 = 0.
The set of equations (4.6) are then
(x − a)2 + y 2 − z 2 − b2 = 0, −2(x − a) = 0, −2b = 0,
which implies that the straight lines defined by
z = ±y
are the envelopes •
Let us see first some particular cases where Eq. (4.1) becomes simple to
solve.
Case (a): Suppose that the PDEFO is of the form
F (p, q) = 0 . (4.8a)
This expression implies that
p = φ(q). (4.8b)
Making q = a, where a in an arbitrary constant, we get from Eq. (4.8a)
p = φ(a),
then, it follows that

dz = pdx + qdy = φ(a)dx + ady,
which easily can be integrated, obtaining
z = φ(a)x + ay + b. (4.9)
Example 4.2. Find the integral surface of the following equation
2 2
∂z ∂z
− = 0.
∂x ∂y
Making q = ∂z/∂y = a, we have that p = ∂z/∂x = ±a, then the solution is
z = ±ax + ay + b •
Case (b): Suppose that the PDEFO can be written in the form
f1 (x, p) = f2 (y, q). (4.10)
Making
f1 (x, p) = f2 (y, q) = a, (4.11)
where “a” is an arbitrary constant, and if we can solve Eq. (4.11) with
respect to p and q, that is
p = φ1 (x, a)
and
q = φ2 (y, a),
then we have
dz = pdx + qdy = φ1 (x, a)dx + φ2 (y, a)dy
which can be integrated, giving the solution
Z Z
z = φ1 (x, a)dx + φ2 (y, a)dy + b. (4.12)
Example 4.3. Find the integral surfaces of the equation

3 2
∂z ∂z
y2 − x3 = 0.
∂x ∂y
This equation can be written in the form (4.10), then we can make
3 2
−3 ∂z −2 ∂z
x =y = a,
∂x ∂x
and from these relations, we have
p = a1/3 x, q = ±a1/2 y.
Then the solution of Eq. (4.12) is
1 a1/2 2
z = a1/3 x2 ± y +b •
2 2
Case (c): Suppose that the PDEFO is of the form
F (z, p, q) = 0. (4.13)
Making z = z(u) where u = ax+y, Eq. (4.13) is transformed to an ordinary

differential equation,

dz dz
F z, a , = 0, (4.14)
du du
which can be integrated, obtain a solution of the form
z = Φ(ax + y, a, b).

∂z ∂z
z= .
∂x ∂y
The new variable u = ax + y transform this equation to
2
dz
z=a .
du
Integrating this equation brings about the solution
1 1/2 2
z= a x + a−1/2 y + b • (4.15)
4
Case d): If the PDEFO can be written in the form
z = px + qy + f (q, p), (4.16)
called Clairaut’s equation, we can make p = a, q = b, having the differential
dz = adx + bdy,
z = ax + by + f (a, b). (4.17)

2
∂z ∂z ∂z ∂z
z=x +y + .
∂x ∂y ∂x ∂y
The equation is of the form (4.16), then the solution is
z = ax + by + a2 b •
Let us see now the Lagrange-Charpit method for the solution of the PDEFO
given by Eq. (4.1). Suppose we find a function U (x, y, z, p, q) such that,
from the equations
F (x, y, z, p, q) = 0 (4.18a)
and
U (x, y, z, p, q) = a, (4.18b)
where “a” ia an arbitrary constant and ∂(F, U )/∂(p, q) 6= 0, we can have
p = p(x, y, z, a) (4.19a)
and
q = q(x, y, z, a). (4.19b)
Therefore we need only to integrate
dz = p(x, y, z, a)dx + q(x, y, z, a)dy (4.20)
to obtain the two parameters solution
Φ(x, y, z, a, b) = 0. (4.21)
The function U is determined from the integrability condition of the Pfaf-
fian equation (4.20),which defines the vector field E = (p, q, −1), and this
integrability (E · ∇ × E = 0), as it was already seen in Eq. (1.54), is given
by
∂q ∂p ∂p ∂q
p −q − + = 0. (4.22)
∂z ∂z ∂y ∂x
Thus, if we can find a function U satisfying Eq. (4.22), assuming we also
have relations (4.19), the integration of Eq. (4.20) can be made obtaining
the solution (4.21). As it was already pointed out, the existence of relations
(4.19) can be assured by asking that the following relation be satisfied
∂(F, U )
6= 0 (4.23)
∂(p, q)
and by the implicit function theorem (see reference T.M. Apostol). Let
us calculate the partial derivatives ∂q/∂z, ∂p/∂z, ∂p/∂y and ∂q/∂x using
relations (4.18). Doing the total derivation with respect x of Eq. (4.18), we
have
∂F ∂F ∂p ∂F ∂q
+ + =0
∂x ∂p ∂x ∂q ∂x
and
∂U ∂U ∂p ∂U ∂q
+ + =0
∂x ∂p ∂x ∂q ∂x
which can be written as

Fp Fq ∂p/∂x Fx
=− .
Up Uq ∂q/∂x Ux
Due to conditions (4.23), the matrix can be inverted. Solving this equation,
if follows
 
    ∂(U, F ) ∂(U, F )
∂p/∂x Uq Fx −Fq Ux − /
 = −1   =  ∂(q, x) ∂(q, p) .

∂(F, U )/∂(p, q)  ∂(U, F ) ∂(U, F ) 
∂q/∂x −Up Fx Fp Ux /
∂(p, x) ∂(q, p)
(4.24)
In a similar way, after making the total derivation of Eq. (4.18) with respect
to y and z (the index ”x” changes to ”y” and ”z”), we obtain
∂p ∂(F, U ) ∂(F, U )
=− / , (4.25)
∂y ∂(y, q) ∂(p, q)
∂p ∂(F, U ) ∂(F, U )
=− / , (4.26a)
∂z ∂(z, q) ∂(p, q)
and
∂q ∂(F, U ) ∂(F, U )
=− / . (4.26b)
∂z ∂(p, z) ∂(p, q)
Substituting Eq. (4.24), Eq. (4.25) and Eq. (4.26) in Eq. (4.22) we get

∂(F, U ) ∂(F, U ) ∂(F, U ) ∂(F, U )
p − −q − + − =0
∂(p, z) ∂(z, q) ∂(y, q) ∂(p, x)
or
∂U ∂U ∂U
Fp + Fq + (pFp + qFq )
∂x ∂y ∂z
∂U ∂U
− (Fx + pFz )
− (Fy + qFz ) = 0, (4.27)
∂p ∂q
which is a homogeneous linear PDEFO. Thus, From the equations for this
characteristic curves
dx dy dz dp dq
= = = = , (4.28)
Fp Fq pFp + qF q −(Fx + pFz ) −(Fy + qFz )
we can derive at least one characteristic surface
U1 (x, y, z, p, q) = a, (4.29)
which we can use as our Eq. (4.18b), if it satisfies

∂(F, U )
6= 0. (4.30)
∂(p, q)
Therefore, once we have found the pair of relations (4.18), we can obtain
Eq. (4.19) and integrating Eq. (4.20), we get the solution Eq. (4.21).

2
∂z ∂z
yz − = 0.
∂x ∂y
The function F is given by
F (x, y, z, p, q) = yzp2 − q = 0, (4.31)
and the partial derivatives are
Fp = 2yzp, Fq = −1, Fz = yp2 , Fx = 0, and Fy = zp2 .
The equation for the characteristic (4.28) are
dx dz dp dq
= −dy = 2 = 3
= .
2pyz 2p yz − q −yp −(zp + yp2 q)
2
From Eq. (4.31),one gets q = yzp2 which can be used in the third term
above, and using the fourth term, we get
dz dp
2
=
yp z −p3 y
U1 = pz = a. (4.32)
With Eq. (4.31) and Eq. (4.32), we obtain
p = a/z
and
q = ya2 /z.
In this way, the Pfaffian equation (4.20) is
a ya2
dz = dx + dy
z z
and can be integrated, giving the solution
z 2 = 2ax + a2 y 2 + b.
Now, the problem of finding the integral surfaces passing through a

given initial curve determined by
x = x(t), y = y(t), and z = z(t), (4.33)
can be solved finding first the function
b = b(a) (4.34)
in such a way that the envelope of the one-parameter family of surfaces
Φ(x, y, z, a, b(a)) = 0 (4.35)
determined by Eq. (4.35) and
∂Φ ∂Φ ∂b
+ =0 (4.36)
∂a ∂b ∂a
passes through the curve define by Eq. (4.33).
The functionality b = b(a) is easier determined by Eq. (4.35) and

∂Φ ∂x ∂Φ ∂y ∂Φ ∂z
+ + = 0. (4.37)
∂x ∂t ∂y ∂t ∂z ∂t
This means that the tangent vector of the initial curve must be orthogonal
to the normal vector of the surface Φ = 0. Once b(a) is determined, this
solution is obtained by finding the enveloped of the one parameter family
of surfaces already deduced.
Example 4.7. Find the integral surface of the Example (4.4) which passes
through the curve
x(t) = 1, y(t) = t and z(t) = t.
The two parameter family of integral surface is given by 4z = (a1/2 x +
a−1/2 y + b)2 , which defines Φ(x, y, z, a, b) = (a1/2 x + a−1/2 y + b)2 − 4z.
Then, the Eq. (4.35) and Eq. (4.37) are given on the initial curve as
2
a1/2 + a−1/2 t + b(a) − 4t = 0
and

2 a1/2 + a−1/2 t + b(a) a−1/2 − 4 = 0.
From where we obtain

a=t, and b(a) = a1/2 − a−1/2 a = 0 . (4.38)
Thus, we have a one-parameter family of integral surfaces given by

2
Φ = a1/2 x + a−1/2 y − 4z = 0. (4.39)
The envelope of these surfaces is determined by Eq. (4.39) and ∂Φ/∂a = 0,

from this one, we obtain
h x y i
a1/2 x + a−1/2 y − = 0. (4.40a)
a1/2 a3/2
From this equation, the value of “a” is determined as
a = y/x,
and substituting this result in Eq. (4.39), we obtain finally the solution
z = yx • (4.40b)
If the system (4.28) can be integrated, then the solve the Cauchy problems
the following method called “the Characteristics”, “Cauchy”, or sometimes
“First Jacobi” method is very useful. Taking s as the parameter for the
family curves Γs
x0 = x0 (s), y0 = y0 (s) and z0 = z0 (s), (4.41)
the integral surface z = z(x, y) passing through the curve can be imagined
as being formed by the set of point the certain monoparametric parameter
family of curves Γ, called characteristic (see Fig. 4.2)
x = x(t, s), y = y(t, s) and z = z(t, s), (4.42)

where is necessary to ask for
∂(x, y)
6= 0 (4.43)
∂(s, t)
to be able to have the inverse relations
s = s(x, y) (4.44a)
and
t = t(x, y). (4.44b)
Let us assume that z = (x, y) is twice differentiable and is the integral
surface of
F (x, y, z, p, q) = 0. (4.45)
Fig. 4.2 Curve of initial conditions crossing the characteristics.
Doing the differentiation of Eq. (4.45) with respect to x and y, we have

∂p ∂q
Fx + pFz + Fp + Fq =0 (4.46a)
∂x ∂x
and
∂p ∂q
Fy + qFz + Fp + Fq = 0. (4.46b)
∂y ∂y
Now, because of equality
∂q ∂p
= ,
∂x ∂y
∂p ∂p
Fx + pFz + Fp + Fq =0 (4.47a)
∂x ∂y
and
∂q ∂q
Fy + qFz + Fp + Fq = 0, (4.47b)
∂x ∂y
which represent linear partial differential equation for p and q. The equation
for the characteristic of the system (4.47), taking p and q as unknown, are
(see 2.16)
dx dy dp dq
= = = = dt, (4.48)
Fp Fq −(Fx + pFZ ) −(Fy + qFZ )
and because z is related with p and q through the expression

dz = pdx + qdy,
then along one of the characteristic of Eq. (4.48), this expression will given
by
dz dx dy
=p +q = pFp + qFq ,
dt dt dt
or
dz
= dt. (4.49)
pFp + qFq
With this last expression, we complete the system
dx dy dz dp dq
= = = = = dt. (4.50)
Fp Fq pFp + qFq −(Fx + pFZ ) −(Fy + qFZ )
From these equation we can obtain the solution
x = x(t), y = y(t), z = z(t), p = p(t) and q = q(t),
that is, we can find the characteristics
x = x(t), y = y(t), and z = z(t).
However, one can not guarantee uniqueness of these equations since one has
only the initial data x0 (s), y0 (s) and z0 (s), one still needs the conditions
p0 (s) and q0 (s). To obtain these conditions, let us see first the behavior of
the integral surface of Eq. (4.45) along the characteristic curves. Calculat-
ing the total variation of Eq. (4.45) along an integral curve of the system
(4.50),
dF dx dy dz dp dq
= Fx + Fy + Fz + Fp + Fq ,
dt dt dt dt dt dt
and using Eq. (4.50), we have
dF
= Fx Fp + Fy Fq + Fz [pFp + qFq ] − Fp [Fx + pFy ] − Fq [Fy + qFz ] = 0
dt
then, F is constant along an integral curve of the system (4.50), that is
F (x, y, z, p, q) = F (x0 , y0 , z0 , p0 , q0 ), (4.51)
and if Eq. (4.50) is satisfied, we must choose the initial values
x0 (s), y0 (s), z0 (s), p0 (s) and q0 (s) such that
F (x0 , y0 , z0 , p0 , q0 ) = 0. (4.52)
This expression give us the first condition to find p0 (s) and q0 (s). To find
the other condition, let us assume that q0 (s) is given. Then, integrating
the system (4.50) such that Eq. (4.52) is satisfied, we obtain
x = x(s, t), y = y(s, t), z = z(s, t), p = p(s, t), q = q(s, t). (4.53a)
Fixing s, we obtain one the characteristic curves
x = x(s, t), y = y(s, t), z = z(s, t), (4.53b)
and taking the variation of s, we obtain a surface. At any point of this
surface, Eq. (4.53) and Eq. (4.45) satisfied. We still need to verify that
∂z ∂z
p= and q =
∂x ∂y
that is, the following expression must be satisfied
dz = pdx + qdy (4.54)
or, in terms of the variable s and t

∂x ∂x ∂y ∂y ∂z ∂z
dz = p ds + dt + q ds + dt = ds + dt
∂s ∂t ∂s ∂t ∂s ∂t
which is equivalent to the following conditions
∂x ∂y ∂z
p +q = (4.55)
∂s ∂s ∂s
and
∂x ∂y ∂z
p
+q = . (4.56)
∂t ∂t ∂t
The Eq. (4.56) is an identity since in Eq. (4.50) we already imposed the
condition (4.54) along the characteristics. However, Eq. (4.55) is not guar-
anteed by any expression. Let us find the conditions to satisfy Eq. (4.55).
Defining the function U as
∂x ∂y ∂z
U =p +q − (4.57)
∂s ∂s ∂s
and differentiating with respect to t, we have
∂U ∂p ∂x ∂2x ∂q ∂y ∂2y ∂2z
= +p + +q − . (4.58)
∂t ∂t ∂s ∂t∂s ∂t ∂s ∂t∂s ∂t∂s
Now, doing the differentiation of Eq. (4.56) with respect to s, we get
∂p ∂x ∂2x ∂q ∂y ∂2y ∂2z
+p + +q − =0
∂s ∂t ∂s∂t ∂s ∂t ∂s∂t ∂s∂t
which can be used in Eq. (4.58) to obtain

∂U ∂p ∂x ∂q ∂y ∂p ∂x ∂q ∂y
= + − − , (4.59)
∂t ∂t ∂s ∂t ∂s ∂s ∂t ∂s ∂t
where we have assumed that the characteristic (4.53b) are twice continu-
ously differentiable. Eq. (4.59) along with Eq. (4.50) takes the form
∂U ∂x ∂y ∂p ∂q
= −[Fx + pFz ] − [Fy + qFz ] − Fp − Fq .
∂t ∂s ∂s ∂s ∂s
Rearranging terms in this equation and adding and subtracting Fz ∂z/∂s,
it follows that

∂U ∂x ∂z ∂p ∂q ∂x ∂y ∂z
= − Fx + Fz + Fp + Fq − Fz p +q −
∂t ∂s ∂s ∂s ∂s ∂s ∂s ∂s
that is
∂U ∂F
=− − Fz U.
∂t ∂s
But, since Eq. (4.1) is satisfied, it follows that
∂U
= −Fz U
∂t
Z t
U (s, t) = U0 (s) exp − Fz dt . (4.60)
0
If the initial conditions are selected such that U0 = 0, we have that U ≡ 0
for any t. Therefore, from the definition (4.57), Eq. (4.55) is satisfied.
Consequently, Eq. (4.54) is also satisfied. The initial conditions for the
functions p and q, p0 (s) and q0 (s), are deduced from
F (x0 , y0 , z0 , p0 , q0 ) = 0 (4.61a)
and
∂x0 ∂y0 ∂z0
p0 + q0 = . (4.61b)
∂s ∂s ∂s
Example 4.8. Find the integral curve of equation

∂z ∂z
z= . (4.62)
∂x ∂y
passing through the line x = 1, z = y.
Let us find first the parametric expressions for the initial curve and the
initial values for p and q. The initial parametric curve Γs is given by
x0 (s) = 1, y0 (s) = s, z0 (s) = s.
Using Eq. (??) we have

s = p0 q0
and
q0 = 1.
The complete initial conditions are
x0 (s) = 1, y0 (s) = s, z0 (s) = s, p0 (s) = s and q0 (s) = 1.
Our function F is given by
F = pq − z,
then the system (4.50) is written as
dx dy dz dp dq
= = = = = dt,
q p 2pq p q
from which the following solution are gotten
p = c1 e t ,
q = c2 e t ,
x = c2 et + d1 ,
y = c1 et + d2
and
z = c1 c2 e2t + d3 ,
where c1 , c2 , c3 , d1 , d2 and d3 are constant. Using the above initial condi-
tions these constant are determined, obtaining the solution
p(s, t) = set ,
q(s, t) = et ,
x(s, t) = et ,
y(s, t) = set
and
z(s, t) = se2t .
Eq. (4.43) is readily satisfied
∂(x, y)
= −et ,
∂(s, t)
and the inverse relations (4.44) are

x
s= ,
y
and
t = log x.
Thus, the integral surface satisfying the initial conditions is given by
z = xy •
Note that this solution is the same as that of Example (4.7), Eq. (4.40b),
which was obtained using a different method. The use of particular method
depends on the skill of the individual.
Let us make the following two observations:
Observation 1. The special cases (a), (b), (c) and (d) presented at the begin-
ning of this chapter can be deduced from the Lagrange-Charpit method by
solving Eq. (4.27) with the corresponding form F of the nonlinear PDEFO.
This is left as exercises at the end of this chapter.
Observation 2. One must note that in the case of having a quasi-linear

(or linear) PDEFO defined in <2 as Eq. (2.15), this one can be expressed
as
F = pP (x, y, z) + qQ(x, y, z) − R(x, y, z) = 0 ,
having the following differentiations
Fx = pPx + qQx − Rx , Fy = pPy + qQy − Ry , Fz = pPz + qQz − Rz ,
Fp = P, Fq = Q .
Therefore, the equations for the characteristics Eq. (4.50) are written for
this case as
dx dy dz −dp
= = =
P (x, y, z) Q(x, y, z) R(x, y, z) pPx + qQx − Rx + p(pPz + qQz − Rz )
−dq
= .
pPy + qQy − Ry + q(pPz + qQz − Rz )
The first three terms on the left side are decoupled from the other terms and
defines the characteristic curves for this type of PDEFO, Eq. (2.16a). Thus,
the projection of the characteristics in the space (x, y, z) of this quasi-linear
PDEFO brings about the characteristic curves studied before for these type
of equations defined in <2 .
4.2. Non-linear PDEFO for Functions Defined in Rn 143
4.2 Non-linear PDEFO for Functions Defined in Rn
The PDEFO of functions defined in n-dimensional space (x1 , ..., xn ) have

the form
F (x, z, p) = 0, (4.63)
where x and p are n-dimensional vectors defined as
x = (x1 , ..., xn ) (4.64a)
and
p = (p1 , ..., pn ), (4.64b)
where pi is given by
∂z
pi = i = 1, ..., n (4.65)
∂xi
and z = z(x) is a n-dimensional surface. It is clear that the Lagrange-
Charpit idea for solving PDEFO defined in <n does not work any more
for n > 2 since the integrability condition of the Pfaffian equation,
Pn
i=1 pi dxi − dz = 0, would imply the coupling of proposed functions
Ui (x, z, p) = ai for obtaining pi = pi (x, z, a). However, one will see that the
Cauchy initial conditions approach will work, in principle, for arbitrary di-
mensional space <n . Cauchy’s problem for Eq. (4.63) consists of determine
the n-dimensional surface passing through the n − 1 dimensional surface Γs
given by
z0 = z0 (s), xi0 = xi0 (s) (4.66a)
where s is a n − 1 dimensional vector
s = (s1 , ..., sn−1 ). (4.66b)
What we have done for two-dimensional space is immediately generalized
to this n-dimensional space. Making the differentiation of Eq. (4.65) whit
respect to the variable xi ,
n
X ∂pl
Fxi + pi Fz + Fp l = 0 i = 1, ..., n, (4.67)
∂xi
l=1
and assuming that the function z is twice continuously differentiable,

∂pl ∂2z ∂pi
= = i, l = 1, ..., n,
∂xi ∂xi ∂xl ∂xl

n
X ∂pi
Fp l = −(Fxi + pi Fz ) i = 1, ..., n. (4.68)
∂xl
l=1
This represents a partial differential equation for pi (i = 1, ..., n), and the
equations for its characteristics curves are given by
dx1 dxn dp1 dpn
= ... = = = ... = = dt (4.69)
Fp1 Fpn −(Fxi + pi Fz ) −(Fxn + pn Fz )
which can be completed by using the fact that along the characteristic
curves, we must have
n n
dz X dxl X
= pl = pl Fpl . (4.70)
dt dt
l=1 l=1
Then, we can written the equation system for the characteristic as
dx1 dxn dz dp1 dpn
= ... = = n = = = dt.
Fp1 Fpn X −(Fx1 + p1 Fz ) −(Fxn + pn Fz )
pl Fpl
l=1
(4.71)
Thus, assuming we know the initial values for pi ,
pi0 = pi0 (s) i = 1, ..., n, (4.72)
and integrating the system (4.71), we can obtain the solution
xi = xi (s, t) i = 1, ..., n, (4.73a)
z = z(s, t), (4.73b)
and
pi = pi (s, t) i = 1, ..., n (4.73c)
which satisfy the initial conditions (4.66a) and (4.72). For a fixed vector s,
the solution (4.73a) and (4.73b) determine a curve in the (n+1)-dimensional
space (x, z), called characteristic. As the vector s changes, we generated
a n-dimensional surface passing through the initial (n − 1)-dimensional
surface Γs . In order for the inverse relation (4.73a) to be satisfied, we have
to ask for the Jacobian of the transformation to be different form zero (see
reference of T.M. Apostol),
 ∂x ∂x1 ∂x1 
1
···
 ∂s1 ∂sn−1 ∂t 
∂(x)  . .. .. 
= det  .
. · · · . .  6= 0.
 (4.73d)
∂(s, t) 
 ∂x
n ∂xn ∂xn 
···
∂s1 ∂sn−1 ∂t
We will show that with a suitable selection of values for pi0 (i = 1, ..., n),
the set of points determined by the characteristics (4.73) form the n-
dimensional integral surface of Eq. (4.63). These values must satisfy
F (x(s, t), z(s, t), p(s, t)) = 0 (4.74)
and
∂z
pi = i = 1, ..., n (4.75a)
∂xi
or equivalently
n
X
dz = pl dxl . (4.75b)
l=1
Differentiating the function F with respect to t, we have
n n
dF X dxi dz X dpi
= Fx i + Fz + Fpi ,
dt i=1
dt dt i=1 dt
and considering this equation along the integral curves of the system (4.71),
we get
n n n
dF X X X
= Fxi Fpi + Fz p i Fp i − Fpi (Fxi + pi Fz ) = 0.
dt i=1 i=1 i=1
Consequently, the function F is constant along these curves, that is
F (x, z, p) = F (x0 , z0 , p0 ). (4.76)
Thus, the values of pi0 (i = 1, ..., n) must be such that
F (x10 , ..., xn0 , z0 , p10 , ..., pn0 ) = 0. (4.77)
We still need to show that the expression
Xn
dz = pi dxi (4.78)
i=1
is satisfied for every t and sj (j = 1, ..., n − 1). Expressing Eq. (4.78) in
terms these parameters, it follows that
 
n−1 n n
∂z X ∂z X ∂xi
X ∂x i
dz = dt + dsj = pi  dt + dsj  ,
∂t j=1
∂sj i=1
∂t j=1
∂sj
and equaling coefficients of the independent variables, we obtain two con-

ditions
n
∂z X ∂xi
− pi =0 (4.79a)
∂t i=1 ∂t
and
n
∂z X ∂xi
− pi = 0. (4.79b)
∂sj i=1
∂sj
Eq. (4.79a) represents an identity along the integral curves of the system
(4.71) since on these curves we have
n
∂z X ∂xi
= p i Fp i and = Fp i i = 1, ..., n.
∂t i=1
∂t
However, Eq. (4.79b) will be true only for certain values of pi0 (i = 1, ..., n).
To see this, define the function Uj as
n
∂z X ∂xi
Uj = − pi j = 1, ..., n − 1, (4.80)
∂sj i=1
∂sj
and carrying out the differentiation of Eq. (4.79a) and Eq. (4.79b) with
respect to parameter t, we get the following two equations
n n
∂Uj ∂2z X ∂ 2 xi X ∂pi ∂xi
= − pi − (4.81a)
∂t ∂t∂sj i=1
∂t∂sj i=1
∂t ∂sj
and
n n
∂2z X ∂ 2 xi X ∂pi ∂xi
− pi − = 0. (4.81b)
∂t∂sj i=1
∂t∂sj i=1
∂sj ∂t
Using Eq. (4.81b) in Eq. (4.81a) and the assumption that x, y, z ∈ C 2 (<2 ),
we obtain for Eq. (4.81a)
n n
∂Uj X ∂pi ∂xi X ∂pi ∂xi
= − . (4.82)
∂t i=1
∂sj ∂t i=1
∂t ∂sj
Evaluating this expression along the integral curves of Eq. (4.71), we have
n n
∂Uj X ∂pi X ∂xi
= Fpi + (Fxi + pi Fz ) .
∂t i=1
∂sj i=1
∂sj
Adding and subtracting the term Fz (∂z/∂sj ), it follows that

n
∂Uj X ∂pi ∂xi ∂z
= Fpi + Fx i + Fz
∂t i=1
∂sj ∂sj ∂sj
" n
#
∂z X ∂xi ∂F
− Fz − pi = − Fz U j .
∂sj i=1
∂s j ∂sj
Taking into consideration Eq. (4.79b), it follows that ∂F/∂sj = 0, and the
solution of the above equation is
Z t
Uj = Uj0 (s) exp − Fz dt . (4.83)
0
In this way, if we want Eq. (4.79b) to be satisfied, we must have that Uj = 0

for any t, and to get this it is enough to have Uj0 = 0 which implies from
Eq. (4.80) that the pi0 (i = 1, ..., n) must be such that
n
∂z0 X ∂xi0
− pi0 = 0 j = 1, ..., n − 1. (4.84)
∂sj i=1
∂sj
Once we have the values pi0 for i = 1, ..., n satisfying the Eq. (4.77) and
Eq. (4.84), the solution (4.73) is completely determined.
Example 4.9. Find the integral surface of the equation

n 2
X ∂z
=4
i=1
∂xi
passing through the (n − 1)-dimensional surface Γs defined by

n−1
X
xn0 = 1, z= xi .
i=1
Let us find first the parametric form of this initial conditions. Our function
F is
n
X
F = p2i − 4 = 0,
i=1
and the (n − 1)-dimensional surface Γs is given by

n−1
X
xn0 = 1, z0 = si , xi0 = si i = 1, ..., n − 1.
i=1
Eq. (4.84) brings about the following relations

n
∂z0 X ∂xi0
− pi0 = 1 − pj0 = 0 j = 1, ..., n − 1
∂sj i=1
∂sj
and
√
pn0 = ± 5 − n.
Now, the system (4.71) has the following form

dxi dxn dz dp1 dpn
= ... = = n = = ... = = dt
2pi 2pn X
2
0 0
2 p1
i=1
which has the solutions
pi = ci i = 1, ..., n,
xi = 2ci t + di i = 1, ..., n,
and !
n
X
z=2 c2i t + d0 .
i=1
Using the initial conditions in these expressions, it follows that
√
cn = ± 5 − n ci = 1 i = 1, ..., n − 1,
dn = 1 di = si i = 1, ..., n − 1,
and
n
X
d0 = si .
i=1
So, the solution of problem is given by
xi (s, t) = 2t + si i = 1, ..., n − 1, (4.85a)
√
xn (s, t) = ±2 5 − nt + 1, (4.85b)
and
n−1
X
z(s, t) = 8t + si . (4.85c)
i=1
From Eq. (4.85a) and Eq. (4.85b), we see that
∂(x)
= 1,
∂(s, t)
and then the inverse relations can be given which are written as
xn − 1
si = xi ± √ i = 1, ..., n − 1,
5−n
and
xn − 1
t = ±√ .
5−n
Thus, using these relations and Eq. (4.85c), the integral surfaces can be
integrated as
n−1
√ X
z(x) = ±(xn − 1) 5 − n + xi .
i=1
Note that real solutions are obtained for this problem for the dimension of
the space lower or equal to five, n ≤ 5.
4.3 Problems
4.1 Find the envelope to the family of surfaces defined by
z = ax + a−1 y − (b − x)2 .
4.2 Find the integral surfaces of the equation

2 2
∂z ∂z ∂z
− = 4.
∂x ∂x ∂y
Suggestion sees the Example (4.2).
4.3 Find the solution of the following equation

3 1/3
2y ∂z −3 ∂z
e −x
∂x ∂y
4.4 Find the solutions of the following equation

∂z ∂z
= 9z 2 .
∂x ∂y

" 2 #
2
2 ∂z 2 ∂z ∂z ∂z
xyz = x y +y x + xy − .
∂x ∂y ∂x ∂y

2
∂z ∂z
yz − = 0.
∂x ∂y
4.7 Find te integral surfaces of the equation

∂z ∂z 1 ∂z ∂z
z=x +y +
∂x ∂y 4 ∂x ∂y
that passes through the curve y = 0, z = x2 .
4.3. Problems 151
4.8 Find the solution of the problem 4.3, i.e. find the integral surface
(using cauchy’s method) of the equation
1
F = xp + yq + pq − z = 0
4
which passes through the curve y = a, z = x2 .

n 2
X ∂z
−z =4
i=1
∂xi
which passes through the (n − 1)-dimensional surface Γs determined by

n−1
X
xn0 = 1, z= xi .
i=1
4.10 Find the solution of the PDEFO defined in <2 by

∂z ∂z 1 ∂z ∂z
z=x +y +
∂x ∂y 4 ∂x ∂y
such that z(x, 0) = x2 .
4.11 Demonstrate that the solution of the PDEFO

∂z ∂z
z=
∂x ∂y
which passes through the initial curve Γo : {x = 1, z = y is given by
z(x, y) = xy.
4.12 Find the envelope solution of the PDEFO

2
∂z ∂z
yz − = 0.
∂x ∂y
4.13 Use the Lagrange-Charpit method to find the integral surface of the
non linear PDEFO defined by
i) F (p, q) = 0 .
ii) F (x, y, p, q) = f1 (x, p) − f2 (y, q) = 0 .
iii) F (z, p, q) = 0 .
iv) F (x, y, z, p, q) = px + qy + f (p, q) − z = 0 .
4.4 References
4.1 L. Elsgoltz, Ecuaciones Diferenciales y Cálculo Variacional, Edi-

ciones Cultura Popular, S.A., 1983. Chap. 5.
4.2 N. Piskunov, Differential and Integral Calculus, MIR Publisher, 1969.

Chap. 13.
4.3 T.M. Apostol, Mathematical Analysis, Addison-Wesley Publishing

Co., 1965. Pages 146-148.
4.4 R. Courant and D. Hilbert, Methods of Mathematical Physics, John

Weley & Sons, N.Y., 1962.
Chapter 5
Physical Applications II
In this chapter we will illustrate the use of the PDEFO in some problems of
Physics. We will not try to explain the full physical concepts or to develop
the full physical consequences of solution of the equations, this is not the
intention of this book, but we will give the proper references where the full
physical understanding of the solution can be found.
5.1 Motion of a Classical Particle
Following the ideas of Sec. 1.6, we will present some solutions for the non-
linear PDEFO appearing in that chapter.
5.1.1 Hamilton-Jacobi equation for a one-dimensional

Harmonic Oscillator
When we solved exercise 6 of (I.5), we found that the Hamiltonian for the
one dimensional harmonic oscillator was given by the expression
p2 1
H= + kq 2 , (5.1)
2m 2
where m is the mass of the particle and k is the spring constant. With this
Hamiltonian, the Hamilton-Jacobi equation is given, using (1.75), by
2
1 ∂S 1 ∂S
+ kq 2 + = 0. (5.2)
2m ∂q 2 ∂t
According to Eq. (4.1), the function F is given by
1 2 1 2
F (t, q, S, St , Sq ) = S + kq + St = 0 . (5.3)
2m q 2
153
154 Physical Applications II
So, the partial differentiations are

Ft = 0 , Fq = kq , Fs = 0 , FSt = 1 , FSq = Sq /m ,
and using these in Eq. (4.50), we have the following system
dq dS dSt dSq
dt = = = = .
Sq /m St + Sq2 /m 0 −kq
From the second and last term, we get the equation
dq dSq
=
Sq /m −kq
and solving this one, we obtain the function U as
Sq2 1
U= + kq 2 = a , (5.4)
2m 2
where a is a constant of integration. From Eq. (5.3) and (5.4), we get
∂(F, U )
= Sq /m ,
∂(St , Sq )
then for every value such that Sq 6= 0, we can have
St = −a (5.5a)
and
1/2
√

2a
Sq = ± km − q2 (5.5b)
k
and integrating the expression
ds = St dt + Sq dq ,
we obtain the solution of Eq. (5.2) as
1/2
√ Z
2a 2
S = −at ± mk −q dq + b . (5.6)
k
The trajectory of the particle can be deduced from Eq. (1.78)
∂S
β= .
∂a
5.1. Motion of a Classical Particle 155
Using Eq. (5.6), we obtain

Z
p dq
β = −t ± m/k 1/2
,
[2a/k − q 2 ]
and after rearranging terms and making the integration, the solution can
be written as
p
q = 2a/k cos ω(t + β) , (5.7a)
where a represents the energy of the particles and ω is the angular frequency
of oscillation given as
p
ω = k/m . (5.7b)
The constants a and β are determined by the initial conditions of the prob-
lem, (q0 , p0 ).
5.1.2 The Lagrangian obtained directly from Hamiltonian

If we have the Hamiltonian of a system in some way, we saw in (I.5) that it
is possible to obtain directly the Lagrangian of the system by integrating
the nonlinear PDEFO given by Eq. (1.67). Suppose that we have the
one dimensional dissipative system of example 6, chap. I, section 5, whose
Hamiltonian is given by
p2
H= exp(−2αq/m) , (5.8)
2m
where m is the mass of the particle and α is the friction constant char-
acteristic of the medium. The form of Eq. (1.67) for this case is written
as
2
exp(−2αx/m) ∂L ∂L
−v + L = 0, (5.9)
2 ∂v ∂v
where we have changed the notation (q → x, and q̇ → v) for convenience.
Defining now p̃ and q̃ as
∂L ∂L
p̃ = and q̃ = , (5.10)
∂x ∂v
the function F (see Eq. (4.1)), is given by
exp(−2αx/m) 2
F (x, v, L, p̃, q̃) = q̃ − v q̃ + L = 0 , (5.11)
2m
and the partial differentiations are

α exp(−2αx/m) 2
Fx = − q̃ ,
m2
Fv = −q̃ ,
FL = 1 ,
Fp̃ = 0 ,
and
exp(−2αx/m)
Fq̃ = q̃ − v .
m
Thus the system (4.50) is written as
dx dv dL
= = =
0 exp(−2αx/m) exp(−2αx/m) 2
q̃ − v q̃ − v q̃
m m
dp̃ dq̃
= = .
α exp(−2αx/m)q̃ 2 − p̃ 0
From the last term of this system, we can immediately obtain the function
q̃ given by
q̃ = a ,
where a is a constant integration. Using this in Eq. (5.11), we get a one-
parameter family of solutions given by
exp(−2αx/m) 2
Φ(x, v, L, a) = − a + va − L = 0 , (5.12)
2m
the envelope of the solution is given by (5.1) itself and the relation
∂Φ exp(−2αx/m)
=− a + v = 0. (5.13)
∂a m
The relation (5.13) give us
a = mve2αx/m (5.14)
and using this in Eq. (5.12), we finally obtain
1
L= mv 2 e2αx/m , (5.15)
2
which is the same we have in expression (1.83). Of course, this same ex-
pression could be obtained by using Eq. (1.65a) to get the relation between
v and p, and then using Eq. (1.64).
5.1.3 Relativistic particle moving in a Coulomb field

The Hamiltonian for a relativistic motion of a charged particle in a elec-
tromagnetic field, characterized by the vector potential A and the scalar
potential φ, is given (see reference L.C. Landau and E.M. Lifshitz) by
h e i1/2
H = m2 c4 + c2 (P − A)2 + eφ, (5.16)
c
where m is the mass of the particle, c is the speed of light, e is the charge
of the particle which is an integer multiple of the electron charge (1.602 ×
10−19 Coulomb) because the charge is quantized, and P is the relativistic
momentum vector of the particle. Using this Hamiltonian in the expression
(1.76), squaring the expression obtained, and rearranging terms, we obtain
the Hamilton-Jacobi equation
2
e 2 1 ∂S
∇S − A − 2 + eφ + κ2 = 0 , (5.17)
c c ∂t
where we have define κ = mc. The Coulomb field produced by a point
charge e0 is characterized by the following expression for the potential A
and φ
A=0 and φ = e0 /r , (5.18)
where r is the distance from the charge e0 to e in spherical coordinates. In

these coordinates, Eq. (5.17) has the following expression for our problem
2 2 2
∂S 1 ∂S 1 ∂S
+ 2 + 2 2 −
∂r r ∂θ r sin θ ∂φ
2
1 ∂S α
− 2 + + κ2 = 0 (5.19)
c ∂t r
where α is given by α = ee0 . Defining the variables p1 , p2 , p3 and p4 as
∂S ∂S ∂S ∂S
p1 = , p2 = , p3 = and p4 = ,
∂r ∂θ ∂φ ∂t
the function F (Eq. (4.63)) is given as
p22 p23
F (r, θ, φ, t, S, p1 , p2 , p3 , p4 ) = p1 + + −
r2 r2 sin2 θ
1 α 2
− 2 p4 + + κ2 = 0 . (5.20)
c r
The partial differentiations of this function are

p2 p2 2 α α
Fr = −2 32 − 2 3 3 2 + 2 p4 + ,
r r sin θ c r r2
p2 cos θ
Fθ = −2 23 3 ,
r sin θ
Fφ = 0 ,
Ft = 0 ,
Fp1 = 2p1 ,
Fp2 = 2p2 /r2 ,
Fp3 = 2p3 /r2 sin2 θ and
Fp4 = −2(p4 + α/r)/c2 .
Using this differentiations in Eq. (4.71), we obtain the system of equations
dr dθ dφ dt
= = = =
2p1 2p2 2p3 /r2 sin2 θ −2(p4 + α/r)/c2
r2
dS
= 2 2 =
2p 2p 2p4
2p21 + 22 + 2 32 − 2 (p4 + α/r)
r r sin θ c
dp1
= =
p22 p23 2 α
2 3 + 2 3 2 − 2 (p4 + α/r) 2
r r sin θ c r
dp2 dp3 dp4
= = = . (5.21)
p23 cos θ 0 0
2 2 3
r sin θ
From the last two terms, we have for p3 and p4
p3 = a1 (5.22a)
and
p4 = a2 . (5.22b)
Using Eq. (5.22a) in Eq. (5.21) and the terms containing dθ and dp2 , we
have the equation
dθ dp2
= ,
2p2 p23 cos θ
2 2 3
r2 r sin θ
whose solution is
a21
p22 + = a23 , (5.22c)
sin2 θ
where a3 is the constant of integration. Substituting Eq. (5.22) in Eq.

(5.21) and taking the terms containing dr and dp1 , we have the equation
dr dp1
=
2p1 a23 2 α
2 3 − 2 (a2 + α/r) 2
r c r
and making the integration, we get
a2 2a2 α 1 α2 1
p21 = − 23 + 2 + 2 2 + a4 , (5.22d)
r c r c r
where a4 is another constant of integration. Using the expression (5.22) in
the differential
dS = p1 dr + p2 dθ + p3 dφ + p4 dt
and integrating, we get the solution of Eq. (5.19) as
Z 2 1/2
a3 2a2 α 1 α2 1
S = a1 φ + a2 t ± − 2 + 2 + 2 2 + a4 dr
r c r c r
Z
± [a23 − a21 / sin2 θ]1/2 dθ + b , (5.23)
where b is an integration constant. The constant a4 is not independent of
the other ones because by substituting p1 , p2 , p3 and p4 in Eq. (5.20), we
get
a2
a4 = 22 − κ2 , (5.24)
c
and we can express Eq. (5.23) then as
1/2
(a2 + α/r)2 a23
Z
2
S = a1 φ + a2 t ± − 2 −κ dr
c2 r
Z
± [a23 − a21 / sin2 θ]1/2 dθ , (5.25)
where a1 represents the energy of the particle and a3 its angular momentum.
The trajectory of the particle is determined from relation (1.78)
∂S
β= . (5.26)
∂a3
The trajectory of the particle is given by the algebraic solution of the ex-
pression
Z Z
a3 dr a3 sin θdθ
β=∓ i1/2 ± . (5.27)
[a3 sin2 θ − a21 ]1/2
2
h 2
a
r2 c12 (a2 + α/r)2 − r23 − κ2
A physical discussion of the solution, for the particular case a1 = 0, can be
found in the same reference mentioned previously in this subsection.
5.1.4 Motion of a test particle in a Schwarchild’s space

If in Eq. (5.17), we make A = 0 and φ = 0, the equation can be written as

µν ∂S ∂S
η + κ2 = 0 , (5.28)
∂xµ ∂xν
where repeated indexes means summation over these (Einstein’s conven-

tion), in cartesian coordinates, xµ for µ = 1, 2, 3, 4 are the components of
the vector x = (x, y, z, ct) and η µν are the components of the 4 × 4 matrix
 
1 0 0 0
0 1 0 0 
(η µν ) = 

0
 (5.29)
0 1 0 
0 0 0 −1
which is the inverse matrix of that one formed with the coefficients of the
pseudo-metric in the four dimensional Minskowsky space (flat space). This
metric is
ds2 = ηµν dxµ dxν = dx2 + dy 2 + dz 2 − c2 dt2 . (5.30)
In this way, Eq. (5.28) is interpreted as the Hamilton-Jacobi equation for

a free test particle moving in the flat space defined by the metric (5.30).
General relativity is a correction of the Newton gravity theory for strong
gravity forces which considers that any mass in space-time, and in fact any
form of energy, causes a deformation of our flat space in such a way that
the metric becomes
ds2 = gµν (x)dxµ dxν , (5.31)
where the coefficients of our metric gµν (x) are functions determined by the
kind of energy, matter, and the symmetry of this one. These coefficients can
be calculated by the Einstein’s equation (see reference C. Møller). Thus,
we have passed from a pseudo-Euclidean space described by Eq. (5.30) to
a pseudo-Riemann space characterized by the metric (5.31). Taking the
inverse matrix (g µν (x)), we can calculate the free motion of a test particle
(it is assumed that the deformation caused in the space by the particle is
negligible) in this space with the equation

∂S ∂S
g µν (x) + κ2 = 0 . (5.32)
∂xµ ∂xν
A solution of the Einstein equation for a centrally symmetric gravitational

field is the Schwarzchild solution, given in spherical coordinates as
rg 2 2
ds2 = − 1 − c dt + r2 (sin2 θdφ2 + dθ2 )
r
rg −1 2
+ 1− dr , (5.33)
r
where rg is called Schwarzchild’s radius and it is given by
2kM
rg = , (5.34)
c2
where k is the gravitational constant (k = 6.67×10−8 cm3 /gm sec2 ), and M
is the mass of the object causing this field. The kind of singularity appearing
in the metric (5.33) when r = rg is called chart singularity because it
depends on the particular coordinates we are using to describe the metric
(in other coordinates, this singularity can disappear). The metric matrix
is given by
(1 − rg /r)−1 0
 
0 0
 0 r2 0 0 
(gµν ) =  , (5.35)
 0 0 r2 sin2 θ 0 
0 0 0 −(1 − rg /r)c2
and its inverse is
eν 0
 
0 0
 0 1/r2 0 0
(g µν ) = 

, (5.36)
 0 0 1/r2 sin2 θ 0 
−ν 2
0 0 0 −e /c
where we have defined
eν = 1 − rg /r . (5.37)
Thus, the Hamilton-Jacobi for the motion of a free test particle moving in
the pseudo-Riemann space defined by the metric (5.33) is given, using Eq.
(5.36) in Eq. (5.32), as
2 2 2 2
∂S 1 ∂S 1 ∂S e−ν ∂S
eν + 2 + 2 2 − 2 +κ2 = 0 . (5.38)
∂r r ∂θ r sin θ ∂φ c ∂t
Defining p1 , p2 , p3 and p4 as

∂S ∂S ∂S ∂S
p1 = , p2 = , p3 = and p4 = ,
∂r ∂θ ∂φ ∂t
we have the function F (Eq. (4.63)) given by

p22 p2 e−ν
F = eν p21 + 2
+ 2 3 2 − 2 p24 + κ2 = 0 . (5.39)
r r sin θ c
rg 2 p22 p23 e−2ν rg 2
Fr = p − 2 − 2 + p ,
r2 1 r3 r3 sin2 θ c2 r2 4
2p2 cos θ
Fθ = − 2 3 3 ,
r sin θ
Fφ = 0 ,
Ft = 0 ,
FS = 0 ,
Fp1 = 2p1 eν ,
Fp2 = 2p2 /r2 ,
Fp3 = 2p3 /r2 sin2 θ
and
Fp4 = −2e−ν p4 /c2 .
Using these expressions and Eq. (4.71), we obtain the system of differential
equations
dr dθ dφ dt
ν
= 2
= 2 = −ν
=
2e p1 2p2 /r 2
2p3 /r sin θ −2e p4 /c2
dS
= ν 2 =
2e p1 + 2p2 /r + 2p3 /r2 sin2 θ − 2e−ν p24 /c2
2 2 2
dp1
= 2 =
p1 rg /r − 2p2 /r − 2p3 /r3 sin2 θ + p24 e−2ν rg /c2 r2
2 2 3 2
dp2 dp3 dp4

= 2 2 3 = = . (5.40)
−2p3 cos θ/r sin θ 0 0
From the last two terms, we get for p3 and p4 the following
p3 = a1 (5.41a)
and
p4 = a2 , (5.41b)
where a1 and a2 are constants. Using Eq. (5.41a) in the terms containing
dp2 and integrating with the help of the second term of Eq. (5.40), we get
a21
p22 + = a23 , (5.41c)
sin2 θ
where a3 is a constant. Now, substituting Eq. (5.41a), Eq. (5.41b) and Eq.
(5.41c) in Eq. (5.39), we obtain
e−ν 2 e−2ν 2
p21 + a − 2 a2 + κ2 e−ν = 0 . (5.41d)
r2 3 c
We can use Eq. (5.41) to integrate the differential 1-form in <4
dS = p1 dr + p2 dθ + p3 dφ + p4 dt ,
obtaining the solution for Eq. (5.38) as
1/2
e−2ν 2 e−ν 2
Z
2 −ν
S=± a − a − κ e dr±
c2 2 r2 3
1/2
a21
Z
± a23 − dθ + a1 φ + a2 t + b , (5.42)
sin2 θ
where a3 and a2 represent the angular momentum and the energy of the
particle. The trajectory of the particle is obtained through the relation
resulting from
∂S
β= ,
∂a3
that is
Z Z
a3 dr a3 sin θdθ
β = ± ± . (5.43)
(a23 sin2 θ − a21 )1/2
h i1/2
a22 a23
r2 c2 − r2 + κ2 eν
A physical discussion of the above solution (for the particular case a1 = 0),
can be found in the last above mentioned reference.
5.1.5 Interaction of a periodic gravitational wave with a

test particle
One of the solution of Einstein’s equations in vacuum represents gravita-
tional waves (see reference of I. Robinson and A. Trautman). The metric
ds2 = dx2 + dy 2 + 2(dt)(dr) − σdt2 , (5.44)
represents one of such solutions. The representation here is given in the
chart coordinate {x, y, r, t}, where r and t are related with the coordinate
space z and time T as
√
r = (z − cT )/ 2 (5.45a)
and
√
r = (z + cT )/ 2 (5.45b)
being c the speed of light (c = 2.99792458 × 108 meter/sec). The function

σ is given by
σ(x, y, t) = 2ω(x2 − y 2 ) cos(νt) , (5.46)
where ω represents the amplitude of the wave, and ν is the wave number.
The matrix associated to this metric is
 
1 0 0 0
0 1 0 0 
(gµν ) = 
0
, (5.47)
0 0 1 
0 0 1 −σ
and its inverse matrix is

 
1 00 0
0 10 0
(g µν ) = 

. (5.48)
0 0σ 1
0 01 0
Using Eq. (5.48) in Eq. (5.32), we obtain the Hamilton-Jacobi equation

for a test particle under interaction with the gravitational wave,
2 2 2
∂S ∂S ∂S ∂S ∂S
+ +2 +σ + κ2 = 0 . (5.49)
∂x ∂y ∂r ∂t ∂r
Defining p1 , p2 , p3 and p4 as

∂S ∂S ∂S ∂S
p1 = , p2 = , p3 = , and p4 = ,
∂x ∂y ∂r ∂t
the function F is given by
F (x, y, r, t, S, p) = p21 + p22 + 2p3 p4 + σp23 + κ2 = 0 . (5.50)

Fx = p23 σx ,
Fy = p23 σy ,
Fr = 0 ,
Ft = p23 σt ,
FS = 0 ,
Fp1 = 2p1 ,
Fp2 = 2p2 ,
Fp3 = 2p4 + 2σp3
and
Fp4 = 2p3 .
Using these expressions in the system of equation (4.71), we have
dx dy dr dt dS
= = = = 2 =
2p1 2p2 2p4 + 2σp3 2p3 2
2p1 + 2p2 + 4p3 p4 + 2σp23
dp1 dp2 dp3 dp4
= 2 = 2 = = . (5.51)
−p3 σx −p3 σy 0 −p23 σt
From the term containing dp3 , it follows
p3 = a = constant . (5.52)
Using this result in the terms of Eq. (5.51) containing dt, dp1 and dp2 , we
obtain the following equations
dx 1
= p1 , (5.53a)
dt a
dp1 a
= − σx , (5.53b)
dt 2
dy 1
= p2 (5.54a)
dt a
and
dp2 a
= − σy . (5.54b)
dt 2
Doing the differentiation of Eq. (5.53a) and Eq. (5.54a) with respect to
the variable t and using Eq. (5.53b), Eq. (5.54b), and the definition (5.46),
we obtain an equivalent system given by
d2 x
+ 2ωx cos(νt) = 0 (5.55)
dt2
and
d2 y
− 2ωy cos(νt) = 0 . (5.56)
dt2
From the terms of Eq. (5.51) having dS and dt, we obtain the equation
dS 1
= [p21 + p22 + 2p3 p4 + σp23 ] ,
dt a
but using Eq. (5.50) and Eq. (5.52) here, we get
dS
= −κ2 /a . (5.57)
dt
From the terms of Eq. (5.51) having dr and dt, we have
1
dr = (p4 + σp3 )dt ,
a
where we can use Eq. (5.50), Eq. (5.52) to integrate and to bring about
the expression
Z Z
1 2 2 2 2
r = − 2 [p1 + p2 + σa + κ ]dt + σdt + r0 . (5.58)
2a
The solution of Eq. (5.55) and Eq. (5.56) gives the evolution variables
x and y as a function of the parameter t. Then, from Eq. (5.46), Eq.
(5.53a) and Eq. (5.54a), the evolution of σ, p1 and p2 as a function of the
parameter t can be known. This allows us to integrate Eq. (5.58) to know
the evolution of r as a function of the parameter t. Finally, the evolution
of the variable z and T can be known using Eq. (5.45a) and Eq. (5.45b),
√
z = (t + r)/ 2 (5.59a)
and
√
cT = (t − r)/ 2 . (5.59b)
From Eq. (5.59b), we can calculate the term dT /dt, and since the compo-
nents of the velocity of the particle are given by
−1
dxi dxi dT
vi = = , (5.60)
dT dt dt
we are able to completely know the behavior of the particle under interac-
tion with this gravitational wave. A physical discussion about this solution
can be found in reference G. López.
5.2. Trajectory of a Ray of Light 167
5.2 Trajectory of a Ray of Light
The Electromagnetic field is governed by Maxwell’s equations. When these

equations are solved for vacuum, some of the solutions are plane waves
which are characterized by the property that their direction of propagation
and amplitude are the same everywhere
f = aei(k·r−ωt+α) , (5.61)
where a is the amplitude, k is the wave vector pointing toward the direction
of propagation, ω is the angular frequency of oscillation, α is a constant
phase, and f represents any component of the electromagnetic field (the
real part of Eq. (5.61) has this physical meaning). This type of wave are
the basis for the subject geometric optics. Wherever the wave is not plane,
but geometric optics is still applicable, we can write
f = aeiψ , (5.62)
where ψ is called the Eikonal and satisfies the equation

µν ∂ψ ∂ψ
η = 0, (5.63)
∂xµ ∂xν
where (η µν ) is the metric (5.29), and xµ has the same meaning as above.
In this way, Eq. (5.63) is the equation for the trajectory of a light ray in
the flat space defined by the metric (5.30). Eq. (5.63)can be generalized to
a pseudo-Riemann space through the metric (5.31). The equation for the
trajectory of a light ray in a non-flat space becomes

µν ∂ψ ∂ψ
g (x) = 0. (5.64)
∂xµ ∂xν
As we can see from Eq. (5.28) and Eq. (5.32), Eq. (5.63) and Eq. (5.64)
represent the Hamilton-Jacobi like equations for the particle with zero mass.
In addition, the following approximation can be made if we still have a well
defined frequency of oscillations at any time,
ψ(x) = Ψ(x, y, z)e−iω(x,y,z)t . (5.65)
However, it is common to use the following equation
2 2 2
∂Ψ ∂Ψ ∂Ψ
+ + = n2 (x, y, z) , (5.66)
∂x ∂y ∂z
where n(x, y, z) is the refraction index, defined as
n(x, y, z) = ω(x, y, z)/c . (5.67)
The solution of Eq. (5.66) provides us with the geometric surface which
makes up the front waves in a non-homogeneous medium characterized by
the refraction index n(x, y, z). Defining pi = ∂Ψ/∂xi for i = x, y, z on this
equation, this one can be written as
F = p21 + p22 + p23 − n2 = 0 ,
which has the following equations for its characteristics
dx dy dz dΨ dp1 dp2 dp3
= = = 2 = = = ,
2p1 2p2 2pz 2n 2nx n 2ny n 2nz n
where we have defined ni = ∂n/∂xi . Note that if n does not depend on
one particular variable, the associated p is automatically a constant (or a
characteristic).
5.2.1 Solution of the Eikonal equation for a refraction in-

dex depending on z
Let us consider the case of stratified medium where the refraction index
depends on the variable z only. Eq. (5.66) is written as
2 2 2
∂Ψ ∂Ψ ∂Ψ
+ + = n2 (z) . (5.68)
∂x ∂y ∂z
The solution of this equation can be easily found by proposing the function
Ψ to be of the form
Ψ(x, y, z) = ax + by + u(z) , (5.69)
where a and b are constant and the function u(z) is unknown. Substituting
Eq. (5.69) in Eq. (5.68) and after integration, it follows
Z z
u(z) = (n2 (ξ) − a2 − b2 )1/2 dξ . (5.70)
0
So, the solution of Eq. (5.68) is

Z z
Ψ(x, y, z) = ax + by + (n2 (ξ) − a2 − b2 )1/2 dξ , (5.71)
0
and the light rays in such a medium are found after taking the partial
differentiations with respect the constant appearing in Eq. (5.71), that is
using Eq. (1.78). These partial differentiations are
Z z
∂Ψ dξ
=x−a 2 2 2 1/2
=α
∂a 0 (n (ξ) − a − b )
5.2. Trajectory of a Ray of Light 169
and Z z
∂Ψ dξ
=y−b 2 (ξ) − a2 − b2 )1/2
=β
∂b 0 (n
or Z z
dξ
x=α+a 2 (ξ) − a2 − b2 )1/2
=α (5.72a)
0 (n
and Z z
dξ
y =β+b =β. (5.72b)
0 (n2 (ξ)
− a2 − b2 )1/2
More discussion on this solution can be seen in reference K.R. Lunerburg
and reference M. Born and E. Wolft.
5.2.2 Solution of the Eikonal equation for a refraction in-

dex radially depending
In this case, we have a cylindrical symmetry, and Eq. (5.66) is given by
2 2 2
∂Ψ 1 ∂Ψ ∂Ψ
+ 2 + = n2 (ρ) . (5.73)
∂ρ ρ ∂θ ∂z
The easiest way to obtain the solution of this equation is proposing a solu-
tion of the form
Ψ(ρ, θ, z) = aθ + bz + R(ρ) , (5.74)
and substituting this in Eq. (5.73). The equation gotten for R(ρ) is readily
integrable, obtaining
Z ρ 1/2
2 a2 2
R(ρ) = ± n (ξ) − 2 − b dξ , (5.75)
ρ0 ξ
and the solution for Eq. (5.73) is
Z ρ 1/2
a2
Ψ(ρ, θ, z) = aθ + bz ± n2 (ξ) − 2 − b2 dξ . (5.76)
ρ0 ξ
The light rays in such a medium are given by (using ∂Ψ/∂a = θ0 and
∂Ψ/∂b = z0 )
Z ρ
dξ
θ − θ0 = ±a 1/2 , (5.77a)
ρ0 2 2
ξ n2 (ξ) − aξ2 − b2
Z ρ
dξ
z − z0 = ±b 1/2 . (5.77b)
ρ0 2
n2 (ξ) − aξ2 − b2
A discussion of this solution can be found in the above last reference (for
b = 0).
5.2.3 Eikonal equation for a refraction index radially de-

pending in sphere
In this case, we have a spherical symmetry, and Eq. (5.66) is given by
2 2 2
∂Ψ 1 ∂Ψ 1 ∂Ψ
+ 2 + 2 2 = n2 (r) . (5.78)
∂r r ∂θ r sin θ ∂φ
Making the definitions p1 = ∂Ψ/∂r, p2 = ∂Ψ/∂θ, and p3 = ∂Ψ/∂φ, this
equation is of the form
p22 p2
F = p21 +
2
+ 2 3 2 − n2 (r) = 0 , (5.79)
r r sin θ
which has the following equations for its characteristics
dr dθ dφ dΨ
= 2
= 2 2 = 2 =
2p1 2p2 /r 2p3 /r sin θ 2n (r)
dp1 dp2 dp3
2 3 2 3 2 = 3 3 =
2p2 /r + 2p3 /r sin θ − 2nr n 2 cos θp3 / sin θ 0
The last term implies that p3 = constant (or characteristic). Using this
with the second and fifth terms and after integration, one gets the constant
(or characteristic)
p23
a = p22 + . (5.80)
sin2 θ
Using this in Eq. (5.79), one gets the relation
a
p21 = n2 (r) − 2 (5.81)
r
In this way, one can already integrate the equation dΨ = p1 dr +p2 dθ +p3 dφ
as
s
p23
Z r Z
a
Ψ=± n2 (r) − 2 dr ± a− dθ + p3 φ + b , (5.82)
r sin2 θ
where b is the constant of integration. Again, the partial differentiation of
this expression with respect to the constants a and p3 will bring about the
trajectory of the ray.
5.3 Equivalent Hamiltonians
As it has been mentioned in section (1.6.1), two Hamiltonian H and H e

are equivalents if there is a generatrix function F (canonical transforma-
tion) which is a non separable solution of a non linear partial differential
5.3. Equivalent Hamiltonians 171
equation(1.100-1.103). This particular application is interesting since it al-

lows us to explore other method of solution for this equation which does not
lead us to a separable form of the characteristics method. Let us consider
the example of to find the generatrix function of the form F1 (x, Q) which
makes the Hamiltonian
p2 1
H= + mω 2 x2 and H e = ωP (5.83)
2m 2
equivalents. The function F1 must be a solution of the non linear PDEFO
(1.100) which is written for 1-D as
2
∂F1 1 ∂F1 1
−ω = + mω 2 x2 . (5.84)
∂Q 2m ∂x 2
Defining π = (∂F1 /∂x) and θ = (∂F1 /∂Q), this non linear PDEFO is
written as
π2 1
F = ωθ + + mω 2 x2 = 0 . (5.85)
2m 2
If one wants to used Clariut-Charpit method to find the solution of this
equation, one would need to solve the linear equation for a function
U = U (x, Q, F1 , π, θ) which would have the following equations for its char-
acteristics
dx dQ dF1 dπ dθ dU
= = 2 = = = , (5.86)
π/m ω π /m + ωθ −mω 2 x 0 0
and because one must have that θ = c is one of the characteristic curves,
dF1 = πdx + θdQ is of separable variable type, that is, it is of the form
F1 (x, Q) = f1 (x) + g1 (Q), which in turns means that p = ∂F1 /∂x =
df1 (x)/dx and P = −∂F1 /∂Q = dg1 (Q)/dQ which does not correspond
to a canonical transformation between H and H.
e
Therefore, let us assume that F1 is of the form F1 (x, Q) = f (x)g(Q) and do

the substitution of this expression in Eq (5.79). Then, Eq. (5.79) is written
in the form
2
dg g 2 df 1
ωf + + mω 2 x2 = 0 . (5.87)
dQ 2m dx 2
By selecting f (x) of the form
f (x) = λx2 , (5.88)
one can make factorization
λ2 2 1

dg
x2 λω + g + mω 2 = 0 . (5.89)
dQ 2m 2
So, for any x 6= 0, one gets the equation

dg λ mω
=− g2 − (5.90)
dQ 2mω 2λ
mω Q
g(Q) = − tan (5.91)
λ 2
Thus, the solution obtained for Eq. (5.79) is given by
Q
F1 (x, Q) = −mωx2 tan • (5.92)
2
Let us see another example. Consider the parameter ”s” (length of a circu-
lar accelerator) as the parameter of evolution of of the transversal motion
of a bunch of charged particle in a circular accelerator. The equivalent
Hamiltonians which describe this transversal motion are given by
1 2 e s) = J ,
p + K(s)x2 , and H(J,

H(x, p, s) = (5.93)
2 β(s)
where the function β satisfies the equation 2ββ 00 − β 02 + 4K(s)β 2 = 4
(β 0 = dβ/ds), and K(s) describes the magnetic elements (lattice) in the
accelerator ring. Let us see this equivalence by looking a generatrix function
of the form F1 (x, Q, s), where p = ∂F1 /∂x, J = −∂F1 /∂Q and F1 must
satisfies the non linear PDEFO
2
1 ∂F1 1 ∂F1 1 ∂F1
− = + K(s)x2 + . (5.94)
β(s) ∂Q 2 ∂x 2 ∂s
Assuming now F1 of the form
F1 (x, Q, s) = λx2 ψ(Q, s) , (5.95)
and substituting this expression into our non linear PDEFO, one gets the
following quasi-linear PDEFO for ψ
1 ∂ψ ∂ψ 1
+ = −2λψ 2 − K(s) . (5.96)
2 ∂Q ∂s 2λ
The equations for its characteristics are given by
dψ
β(s)dQ = ds = . (5.97)
−2λψ 2 − K(s)/2λ
From the first two term, one can get the characteristic
Z
ds
c1 = Q − , (5.98)
β(s)
5.3. Equivalent Hamiltonians 173
and from the last two term, one has the following non linear ordinary dif-
ferential equation
dψ
+ 2λψ 2 = −K(s)/2λ . (5.99)
ds
Let us propose a solution for this equation of the form

1
ψ(Q, s) = f (Q) + g(s) . (5.100)
β(s)
Then, since one has that df /ds = (df /dQ)/β(s), because of the character-
istic c1 and after some arrangements, the substitution of Eq. (5.100) into
Eq. (5.99) brings about the following equation for f (Q),
df
+ 2λf 2 = f (β 0 − 4λg) − 2λg 2 − βg 0 + gβ 0 − β 2 K(s)/2λ . (5.101)
dQ
Choosing the function g(s) of the form g = β 0 /4λ, this equation is written
as
1 β 02 ββ 00

df
+ 2λf 2 = − − β 2 K(s) (5.102)
dQ 2λ 4 2
In this way, choosing now the β function such that it satisfies the equation
β 02 ββ 00
− − β 2 K(s) = −1 , (5.103)
4 2
The equation for f is finally stablished as
df 1
+ 2λf 2 + =0. (5.104)
dQ 2λ
By selecting the λ value as λ = 1/2, the solution of this equation is just
written as
f (Q) = − tan Q . (5.105)
Thus, the generatrix function 5.95 is determined as
x2 β 0 (s)

F1 (x, Q, s) = − tan Q − , (5.106)
2β(s) 2
and the two Hamiltonian are equivalents. •
5.4 Problems
5.1 Find the solution of the Hamilton-Jacobi equation of the one dimen-
sional dissipative system of example 6 (Chap. 1, Sec. 1.6) which has the
Hamiltonian
p2
H= exp(−2αq/m) ,
2m
and find the trajectory of the particle.
5.2 Show, in general, that the equation L = vf (x, v) − H(x, f (x, v)) is
the envelope of the nonlinear partial differential equation

∂L ∂L
v − L = H x, ,
∂v ∂v
where a = f (x, v) is the solution of
∂H(x, a)
= v.
∂a
5.3 The magnetic field produced in the vacuum space between the two
conductor of a coaxial cable can be characterized by the following expression
for the potential A and φ

I
A= θ, 0, 0
2π
and
φ = 0,
where I is the current flowing in the conductors. Find the solution of the
Hamilton-Jacobi equation for a charged particle moving in this field.
5.4 Solve Eq. (5.38) for the case ν = 0.
5.5 Solve Eq. (5.49) for σ = constant.
5.6 Solve Eq. (5.66) in spherical coordinates for a refraction index de-
pending on the radius.
5.4. Problems 175
5.7 Write the PDEFO defined in <5 for the generatrix function F1 (x, Q, s)
(see Eq. (1.100)) which establishes the equivalence between the Hamilto-
nians (”s” is the parameter of evolution of the system)
1 2 1 2
p + K1 (s)x2 + p + K2 (s)y 2

H=
2 1 2 2
and
e = 1 P1 + 1 P2
H
β1 (s) β2 (s)
ii) Propose a solution for this equation of the form
F = x2 f1 (Q1 , s) + y 2 f2 (Q2 , s) ,
and show (using the linearly independence of the variables x and y) that
the functions f1 and f2 satisfy the quasi-linear PDEFO given by
1 ∂fi ∂fi
+ + Ki (s)/2 + 2fi2 = 0 , i = 1, 2
βi ∂Qi ∂s
iii) Show that the function
βi0 (s)

1
fi (Qi , s) = − tan Qi −
2βi (s) 2
(where βi0 = dβi /ds) is a solution of this quasi-linear PDEFO, and the
function βi satisfies the non linear ordinary differential equation
2βi βi00 − (βi0 )2 + 4Ki (s)βi2 = 4 , i = 1, 2 .
5.8 A test particle is moving in the space-time manifold defined by the

de Sitter metric
 
100 0
0 1 0 0  h i−2
(gµν ) = f (xµ )   , f (xµ ) = 1 − λ x2 + y 2 + z 2 − (ct)2 2
0 0 1 0 
0 0 0 −1
Determine the Hamilton-Jacobi equation to find the trajectory of the par-
ticle in this space, and establish the equations for the characteristics.
5.5 References
5.1 H. Goldstein, Classical Mechanics, Addison-Wesley Press 1965. Chap.

9, pages 277-279 and 245-287.
5.2 L.C. Landau and E.M. Lifshitz, The Classical Theory of Fields, Perg-
amon Press, 1971. Cap. 7, pages 46-95.
5.3 C. Møller, Theory of Relativity, Oxford University Press, 1952 . Chap.

IX.
5.4 I. Robinson and A. Trautman, Phys. Tev. Lett., 4, No. 8, (1960)

431.
5.5 G. López, Internal Report IFUG-1988, Apd. Postal E-143, León Gua-
najuato, México.
5.6 R.K. Lunerburg, Mathematical Theory of Optics, University of cali-

fornia Press, 1964. Chap. I ,II.
5.7 M. Born and E. Wolf, Principles of Optics, Pergamon Press, 1980.

Chap. III.
Chapter 6
Characteristic Surfaces of Linear

Partial Differential Equation of
Second Order
In this chapter we will use the PDEFO to find the so called “characteristic
surfaces of a linear partial differential equation of second order,” and to see
the usual classification of these equations. In the theory of Partial Differ-
ential Equation Of Second Order (PDESO), the concept of “characteristic
surfaces” has an important role in understanding the type of solution and
its characteristics of existence.
6.1 Characteristic Surfaces of a Linear PDESO Defined in

Rn
For a linear PDESO defined in Rn , one understands a expression of the

form
n
X ∂2u
aij (x) = f (x, u, ∇u), (6.1)
i,j=1
∂xi ∂xj
where x is a point in Ω ⊂ Rn , x = (x1 , ..., xn ), aij is a once continuously

differentiable function defined in Ω, i, j = 1, ..., n. One will assume the
symmetry aij = aji for i 6= j. The function u = u(x) will be at least twice
continuously differentiable in Ω. The function f is defined R2n+1 , and ∇u
represents the expression ∇u = (u1 , ..., un ) with ui = ∂u/∂xi . One can
make the following change of coordinates
ξi = ξi (x) i = 1, ..., n, (6.2)
where one needs to ask for the Jacobian of this transformation to be differ-
ent from zero in Ω
∂(ξ1 , ..., ξn )
6= 0 (6.3)
∂(x1 , ..., xn )
177
178 Characteristic Surfaces of Linear Partial Differential Equation of Second Order
With these variables, it follows that

n
∂ X ∂ξk ∂
= (6.4a)
∂xi ∂xi ∂ξk
k=1
and
n n
∂2 X ∂ 2 ξk X ∂ξk ∂ξl ∂ 2
= + (6.4b)
∂xj ∂xi ∂xj ∂xi ∂xi ∂xj ∂ξl ∂ξk
k=1 l,k=1
Therefore, Eq. (6.1) is transformed to the expression

n n n n
X X ∂ 2 ξk X X ∂ξk ∂ξl ∂ 2 ũ
aij (x) + aij (x)
i,j=1
∂xj ∂xi i,j=1 ∂xi ∂xj ∂ξl ∂ξk
k=1 l,k=1
n n
!
X ∂ξk ∂ ũ X ∂ξk ∂ ũ
=f ξ, ũ, , ..., , (6.5)
∂x1 ∂ξk ∂xn ∂ξk
k=1 k=1
where ũ is defined as
ũ = u(x(ξ)).
Eq. (6.5), can be written as
n
X ∂ 2 ũ
αlk (ξ) = g(ξ, ũ, ∇ũ), (6.6)
∂ξl ∂ξk
l,k=1
where the functions αlk and g have been defined as

n
X ∂ξk ∂ξl
αlk = aij (x) (6.7a)
i,j=1
∂xi ∂xj
and
n n
X ∂ξk ∂ ũ X ∂ξk ∂ ũ
g(ξ, ũ, ∇ξ ũ) = f ξ, ũ, , ..., −
∂x1 ∂ξk ∂xn ∂ξk
k=1 k=1
n n
X X ∂ 2 ξk
− aij (x) . (6.7b)
i,j=1
∂xj ∂xi
k=1
Now, due to (6.3) it is possible to choose the new variables such that all
the elements on the diagonal of the matrix (αlk ) to be nulls, that is, αll = 0
for l = 1, ..., n. From Eq. (6.7a), all of these elements satisfy the same
nonlinear PDEFO,
n
X ∂ω ∂ω
aij (x) = 0, (6.8)
i,j=1
∂xi ∂xj
6.1. Characteristic Surfaces of a Linear PDESO Defined in Rn 179
where ω = ω(x). the solutions of this equation are called “characteristic

surfaces“ of (6.1). These n-surfaces, at the same time, can be used as the
new set of variables (6.2). According to chapter 6(section 2), Eq. (6.8)
represents the following PDEFO
n
X
F (x, ω, p) = aij (x)pi pj = 0, (6.9)
i,j=1
which can be solved by the characteristics method shown in chapter IV.

The equations for the characteristics of (6.9) are
dx1 dxn dω
n = ... = n = n =
X X X
2 a1i (x)pi 2 ani pi 2 aij (x)pi pj
i=1 i=1 i,j=1
dp1 dpn
= n = ... = n . (6.10)
X ∂aij X ∂aij
− pi pj − pi pj
i,j=1
∂x1 i,j=1
∂xn
Example 6.1. Let us find the characteristic surfaces of the equation
3
X ∂2u 1 ∂2u
− = 0. (6.11)
i=1
∂x2i c2 ∂t2
Defining the variable x0 = ct, one has the following dependence u =

u(x) where x = (x1 , x2 , x3 , x0 ). In this case one has
aij (x) = 0, for i 6= j, and i, j = 0, 1, 2, 3. (6.12a)
aij (x) = 1, for i = 1, 2, 3 and a00 (x) = −1. (6.12b)
According to (6.7a) and (6.8), the equation for the characteristic sur-
faces is given by
∂ω 2 ∂ω 2 ∂ω 2 ∂ω 2
+ + − = 0. (6.13)
∂x1 ∂x2 ∂x3 ∂x0
This nonlinear PDEFO is written as
F (x, ω, p) = p21 + p22 + p23 − p20 = 0, (6.14)

and the equations for the characteristics are

dx1 dx2 dx3 dx0 dω
= = = = =
2p1 2p2 2p3 −2p0 2(p21 + p22 + p23 − p20 )
dp1 dp2 dp3 dp0
= = = = .
0 0 0 0
These equations imply that
p1 = b1 , p2 = b2 , p3 = b3 and p0 = b0 , (6.15)
where bj is a constant for j = 0, 1, 2, 3, 4. Eq. (6.14) gives the following
relation among these constants
q
b0 = ± b21 + b22 + b23 . (6.16)
Since one has the relation
dω = p1 dx1 + p2 dx2 + p3 dx3 + p0 dx0 , (6.17)
one gets the solution
q
ω(x) = b1 x1 + b2 x2 + b3 x3 ± b21 + b22 + b23 x0 + b4 (6.18)
Of course, the envelope of this family of surfaces is also a characteristic
surface which can
p be calculated as shown in Chap. 5. Defining Φ = b1 x1 +
b2 x2 + b3 x3 ± b21 + b22 + b23 x0 + b4 − ω and doing Φbi = 0 for i = 1, 2, 3,
one gets the so called ”light cone” as enveloped for this equation (making
b4 = 0 and absorbing the constants in the definition of ω)
ω(x) = x21 + x22 + x23 − x20 .
6.2 Characteristic Surfaces of a Linear PDESO Defined in

R2
Let us see what happens for linear PDESO defined in some domain Ω ⊂ R2 .
In this case, one gets for (6.1)
∂2 ∂2u ∂2u
a11 (x, y) + 2a12 (x, y) + a22 (x, y) = f (x, y, u, ux , uy ). (6.19)
∂x2 ∂x∂y ∂y 2
The new variables
ξ = ξ(x, y), η = η(x, y) (6.20)

bring about the changes
6.2. Characteristic Surfaces of a Linear PDESO Defined in R2 181
∂ ∂ ∂
= ξx + ηx , (6.21a)
∂x ∂ξ ∂η
∂ ∂ ∂
= ξy + ηy , (6.21b)
∂y ∂ξ ∂η
∂2 ∂ ∂ ∂2 ∂2 ∂2
2
= ξxx + ηxx + ξx2 2 + 2ξx ηx + ηxx 2 , (6.22a)
∂x ∂ξ ∂η ∂ξ ∂ξ∂η ∂η
∂2 ∂ ∂ ∂2 ∂2 ∂2
2
= ξyy + ηyy + ξy2 2 + 2ξy ηy + ηyy 2 (6.22b)
∂y ∂ξ ∂η ∂ξ ∂ξ∂η ∂η
and
∂2 ∂ ∂ ∂2 ∂2 ∂2
= ξxy +ηyx +ξx ξy 2 +(ξx ηy +ξy ηx ) +ηx ηy 2 . (6.22c)
∂x∂y ∂ξ ∂η ∂ξ ∂ξ∂η ∂η
Eq. (6.19) can be written in terms of ξ and η as
∂ 2 ũ ∂ 2 ũ ∂ 2 ũ
α1 (ξ, η) 2
+ 2β(ξ, η) + α2 (ξ, η) 2 = g(ξ, η, ũ, ũξ , ũη ), (6.23)
∂ξ ∂η∂ξ ∂η
where the functions α1 , β, α2 , and g are defined as
α1 = a11 ξx2 + 2a12 ξx ξy + a22 ξyy (6.24a)
α2 = a11 ηx2 + 2a12 ηx ηy + a22 ηyy (6.24b)
β = a11 ξx ηx + a12 (ξx ηy + ξy ηx ) + a22 ξy ηy (6.24c)

and
g = f˜(x(ξ, η), y(ξ, η), ũ, ξx ũξ + ηx ũη , ξy ũξ + ηy ũη ) − a11 (ξxx ũξ + ηxx ũη )−
− 2a12 (ξxy ũξ + ηxy ũη )a22 (ξyy ũξ + ηyy ũη ). (6.24d)
One can select α1 = α2 = 0, which brings about, for the functions ξ and η,
the following nonlinear PDEFO
2 2
∂ω ∂ω ∂ω ∂ω
a11 (x, y) +2a12 (x, y) +a22 (x, y) = 0. (6.25)
∂x ∂x ∂y ∂y
This equation cab be seen as a quadratic algebraic equation for the term
ωx (or for ωy ) whose roots are
p
(±) a12 ± a212 − a11 a22
ωx = − ωy . (6.26)
a11
This expression defines a linear PDEFO given by

∂ω ∂ω
+ λ(±) (x, y) = 0, (6.27a)
∂x ∂y
where the function λ(±) is defined as
p
(±) a12 ± a212 − a11 a22
λ (x, y) = . (6.27b)
a11
From (6.23), the resulting linear PDESO is called standard expression
of the linear PDESO (6.19) and is given by
∂ 2 ũ
2β(ξ, η) = g(ξ, η, ũ, ũξ , ũη ). (6.28)
∂η∂ξ
Eqs. (6.27) are used to classify PDESO (6.19) in the following way:
a) if a212 > a11 a22 , the roots λ+ and λ− are reals and different, there-
fore there are two real characteristics curves. This is called hyperbolic case
and one says that Eq. (6.19) is hyperbolic.
b) if a212 = a11 a22 , the roots λ+ and λ− have the same expression (a12 /a11 )
which is real, therefor there is only one characteristic curve. This is called
parabolic case and one says that Eq. (6.19) is parabolic.
c) if a212 < a11 a22 , the roots λ+ and λ− are complex and λ+ = (λ− )∗ ,
therefor, there is not real characteristics. This is called elliptic case and
one says that Eq. (6.19) is elliptic.
Example 6.2. Let us find the characteristic surfaces of the equation

∂2u ∂2u
− x = 0. (6.29)
∂x2 ∂y 2
In this case, one has the following functions
a11 = 1, a12 = 0, a22 = −x.
From, Eq. (6.27b), it follows that
√ √
λ+ = x, λ− = − x,
The partial differential equation for the characteristics surfaces is
(Eq. (6.27a))
∂ω √ ∂ω
± x = 0.
∂x ∂y
6.2. Characteristic Surfaces of a Linear PDESO Defined in R2 183
The equations for the characteristics curves are

dy dw
dx = √ =
± x 0
From these equations one gets two characteristics curves which, at the same
time, are the characteristics surfaces of (6.29). These characteristic surfaces
are
2 2
ξ = y − x3/2 and η = y + x3/2 ,
3 3
and they form the new coordinates for the transformation which leads to
the standard form of Eq. (6.29). The Jacobian of the transformation is
given by
∂(ξ, η) √
= 2 x.
∂(x, y)
Therefor, for x 6= 0, one gets Eq. (6.29) standard form
∂ 2 ũ ũη − ũξ
= .
∂ξ∂η 3(η − ξ)
Note that for x > 0 Eq. (6.29) is of hyperbolic type; for x = 0 Eq.(6.29) is
of parabolic type; and for x < 0 Eq. (6.29) is of elliptical type.
6.3 Problems
6.1 find the characteristic surfaces of the second order linear equation
3
X ∂2u 1 ∂u
− = 0,
i=1
∂x2i D ∂t
where D is constant.
6.2 Find the characteristic surfaces of the equation

∂2u ∂2u
x − 2 = 0.
∂x2 ∂y
6.3 Find the characteristic surface and their standard form of the equa-
tions
∂2u ∂2u
i) − 2 =0.
∂x2 ∂y
∂2u ∂2u
ii) + 2 =0.
∂x2 ∂y
∂ 2 u ∂u
iii) − =0.
∂x2 ∂y
6.4 Determine the regions of the real plane <2 where the following PDESO
∂2u ∂2u
x 2
−y 2 =0
∂x ∂y
is hyperbolic, elliptic, or parabolic, and find the characteristic surfaces and
the standard form for the hyperbolic case.
6.4. References 185
6.4 References
6.1 F.H. Miller, Partial Differential Equations, John Wiley & Sons, 1941.
Chap. II.
6.2 R. Courant and D. Hilbert, Methods of Mathematical Physics, Vol.

II, John Wiley & Sons, 1962. Chap. I, II
6.3 V.S. Vladimirov, Equations of Mathematical Physics, Marcel Dekker,

Inc. N.Y., 1971.

Partial Differential Equations of First Order and Their Applications To Physics G Lopez PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Partial Differential Equations of First Order and Their Applications To Physics G Lopez PDF

Uploaded by

Copyright:

Available Formats

January 22, 2014 14:45 World Scientific Book - 9in x 6in ws-book9x6˙x˙x

I want to dedicate this book to my parents, J. Guadalupe López López and

I also want to dedicate this book to my sister Esperanza and my friend

I finally want to dedicate this book to my granddaughter Jezen for being

This book is a collection of notes of the course Partial Differential Equa-

In this book I try to point out the mathematical importance of Partial

plications of this branch in Physics , and to give to physicists and applied

1. Geometric Concepts and Generalities 1

2. Partial Differential Equations of First Order 33

3.4 Classical Statistical Mechanics in Equilibrium . . . . . . . 71

4. Nonlinear Partial Differential Equations of First Order 127

5. Physical Applications II 153

5.2.1 Solution of the Eikonal equation for a refraction

6. Characteristic Surfaces of Linear Partial Differential

Geometric Concepts and Generalities

1.1 Surfaces and Curves in Three Dimensions

By a surface S in R3 we mean any relation between the rectangular carte-

2 Geometric Concepts and Generalities

Fig. 1.1 Surface of sphere is not unique parameterized.

where g(v) is such that 1 + g(v) > 0 for all v ∈ R. By a curve Γ in R3 we

P = (x(s), y(s), z(s)),

1.1. Surfaces and Curves in Three Dimensions 3

Fig. 1.3 Surface of sphere is generated by the planes z = k, −a ≤ k ≤ a.

be three points on Γ (see Fig. 1.4) . If δc is the Euclidean distance between

4 Geometric Concepts and Generalities

Fig. 1.4 Distances δc and δs between the points P and Q.

the chord P Q are

1.1. Surfaces and Curves in Three Dimensions 5

Fig. 1.5 Part of a curve Γ laying on the sphere.

6 Geometric Concepts and Generalities

on the plane π1 , must be perpendicular to the normal directions ∇F at P ,

1.1. Surfaces and Curves in Three Dimensions 7

and therefore is proportional to this vector, establishing the following equa-

8 Geometric Concepts and Generalities

∇F = (2x, 2y, 2z), the vector normal of the surface F at P is

1.2. Parallelism of vector fields 9

where Y , Z are coordinates of the plane π1 and

or writing these equations in parametric way, it follows

1.2 Parallelism of vector fields

Definition 1.1. Let a, b be two vector fields continuously defined in region

a(P ) = λ(P )b(P ) , (1.20)

10 Geometric Concepts and Generalities

Assume that a = (a1 , a2 , a3 ) and b = (b1 , b2 , b3 ), then, the above rela-

1.3. Methods of Solution of dx/P = dy/Q = dz/R 11

1.3 Methods of Solution of dx/P = dy/Q = dz/R

12 Geometric Concepts and Generalities

Then, with (1.28) we would have that

Example 1.2. Find the integral curves of the equation

METHOD (II). Suppose we find two vectors fields E0 = (P 0 , Q0 , R0 )

1.3. Methods of Solution of dx/P = dy/Q = dz/R 13

to be an exact differential, where ρ is another constant. From Eq. (ref1.26),

14 Geometric Concepts and Generalities

this equation has a solution of the form

where c1 is the integration constant. Solving this equation for z (z =

may be readily obtained.

Example 1.3. Find the integral curves of the equations

1.4. Orthogonal Trajectories of a System of Curves on a Surface 15

1.4 Orthogonal Trajectories of a System of Curves on a

The problem can be set as follows. Given a surface S defined by

16 Geometric Concepts and Generalities

Example 1.4. Find the orthogonal trajectories on the sphere x2 +y 2 +z 2 =

1.5. Pfaffian Differential Equation in R3 17

1.5 Pfaffian Differential Equation in R3