Lectures 4-9 Calculus of Variations
September 28, 2023
Calculus of Variations
Functionals: A functional is a generalisation of a function.
Z b dy
J[y ] = F x, y , dx. (1)
a dx
A function f acts on a number x to give another number f (x).
A functional J acts on a function y (x) to give a number
J[y (x)]. Example of a functional
Z b
J[y ] = (y (x)2 + y 0 (x)y 00 (x))dx
a
J[y] is independent of x, but it is a function of y(x).
EXAMPLE OF A FUNCTIONAL
Z 1
2
h i
0 2 2 2
J[y] = (y (x)) − π (y(x)) dx
0
Different functions y(x) give different values to J[y].
We will be mainly interested in one which gives a
minimum value to J[y ].
Calculus of Variations
The Calculus of Variations is concerned with solving Extremal
Problems for a Functional. We will mainly find a minmising
function for a functional.
Just as the vanishing of the gradient of a function of several
variables determines the critical points.
A functional gradient distinguishes the possible functions that
might be minimizers (maximizers) of the functional.
Vanishing the gradient for a function leads to a system of
algebraic equations for the critical points.
The functional analog results in a boundary value problem
for a nonlinear ordinary or partial differential equation
whose solutions are the minimising functions for the
functional.
Today’s Lecture
Today we will derive a necessary condition for a
functional to have a minimum (maximum) value.
We will see that this will be a second order ordinary
differential equation, called the Euler-Lagrange equation,
with corresponding boundary conditions.
In the following lectures we will see that many problems
of both mathematical and practical importance reduce
to the solution of the Euler-Lagrange equation.
BASIC CONCEPTS
For function to each value of For functional to each function,
the independent variable, there there is a corresponding value
is a corresponding value of the Z 1
dependent variable. J[y ] = y (x)dx
0
y (x) = 2x 3 + 3x. Z 1
= (2x 3 + 3x)dx = 2
The increment of a function ∆y 0
for ∆x increment of the The increment of a functional
argument is ∆J for δy increment of the
argument is
∆y = y (x + ∆x) − y (x)
∆J = J[y + δy ] − J[y ].
BASIC CONCEPTS
(b) Functional and its
(a) Function and its increment,
increment,
∆f = f (x + ∆x) − f (x)
∆J = J[y + δy ] − J[y ].
BASIC CONCEPTS
Differential of a function
∆f (x0 ) = f (x0 + ∆x) − f (x0 ) =
! !
df 1 d 2f
∆x + (∆x)2 +... = df + d 2 f + ...
dx x0 2! dx 2 x0
| {z } | {z }
df : First deviation d 2 f : Second deviation
Variation of a functional
∂2J
∂J 1
∆J[y ] = J[y (x)+δy (x)]−J[y (x)] = δy + (δy )2 +
∂y 2! ∂y 2
| {z } | {z }
δJ: First variation δ 2 J: Second variation
= δJ + δ 2 J + ...
∂J
δJ = δy
∂y
BASIC CONCEPTS
A functional has a local minimum at y0 (x) if there exists a ε > 0
such that for all functions y (x) ∈ C 1 |y (x) − y0 (x)| < ε
∆J = J[y ] − J[y0 ] ≥ 0 (2)
a local maximum if ∆J = J[y ] − J[y0 ] ≤ 0.
Note: If the above relationships are satisfied for arbitrarily large
ε > 0, then the functional has a global minimum or maximum.
Extremising a Functional
The basic minimization problem
Find a function y = y (x) ∈ C 1 [a, b] that minimizes the functional
Z b
J[y ] = F (x, y , y 0 )dx.
a
F (x, y , y 0 ) - called Lagrangian for the variational problem, a
reasonably smooth function of all three of its arguments x, y , y 0 .
In order to uniquely specify a minimizing function, we need to
impose suitable boundary conditions.
We will consider the Dirichlet boundary conditions
y (a) = α, y (b) = β.
Assume y0 (x) ∈ C 1 [a, b] is a minimum value for J(y ) and consider
the family of functions u(x, ε) = y0 (x) + εv ,
ε is a parameter and v (x) is an arbitrary added curve v (x) ∈ C 1
with v (a) = 0, v (b) = 0.
y0 +ϵν(x)
variation ϵν(x)
y0 minimum
a b
x
Extremising a Functional
For fixed v (x) consider
Z b
J[y0 +εv ] = Φ(ε) = F (x, y0 +εv , y00 +εv 0 )dx is a function of ε.
a
We will show that the necessary condition for the extremum of the
functional J[y ] is:
Z b
0 d d
Φ (0) = J[y0 +εv ]/ε=0 = F (x, y0 +εv , y00 +εv 0 )dx/ε=0 = 0.
dε a dε
Fundamental lemma of calculus of variations
If f (x) is continuous in [a, b] and if
Z b
f (x)v (x)dx = 0
a
for every v (x) ∈ C 0 (a, b) such that v (a) = 0 and v (b) = 0,
then f (x) = 0 for all x ∈ [a, b].
The Euler-Lagrange Equation
In view of the fundamental lemma it follows that
∂F d ∂F
− = 0. (3)
∂y dx ∂y 0
This is a second order ordinary differential equation
∂F ∂2F 2
0 ∂ F
2
00 ∂ F
E (x, y , y 0 , y 00 ) = − − y − y = 0, (4)
∂y ∂x∂y 0 ∂y ∂y 0 ∂y 0 2
known as the Euler-Lagrange equation associated with our
variational problem.
Theorem
If the function F (x, y , y 0 ) is at least twice continuously
differentiable with respect to all its arguments:
F (x, y , y 0 ) ∈ C 2 .
Then any minimizer y (x) ∈ C 1 [a, b] to the corresponding
functional Z b
J[y ] = F (x, y , y 0 )dx,
a
subject to the selected boundary conditions, must satisfy the
associated Euler-Lagrange equation.
Second Derivative Test,
Legendre condition
Satisfying the Euler-Lagrange equation does not guarantee
that a function is a minimizer. Any critical function can also
be a maximizer for the functional. The nature of a critical
function will be elucidated by the second derivative test.
In order for y (x) to lead to a minimum (or maximum) of the
functional J[y ], the condition
∂2F
>0 (< 0 maximum) (5)
∂y 02
needs to be fulfilled for this function at all a ≤ x ≤ b.
Extremising a Functional
If y = y (x) ∈ C 1 [a, b] minimizes the functional
Z b
J[y ] = F (x, y , y 0 )dx, (6)
a
with y (a) = α, y (b) = β then F(x,y,y’) satisfies the
Euler-Lagrange equation
∂F d ∂F
− = 0. (7)
∂y dx ∂y 0
Equation (7) is a second order ordinary differential equation
∂F ∂2F 2
0 ∂ F
2
00 ∂ F
E (x, y , y 0 , y 00 ) = − − y − y = 0, (8)
∂y ∂x∂y 0 ∂y ∂y 0 ∂y 0 2
(F (x, y , y 0 ) is known as the Lagrangian for the variational
problem)
Special cases of the Euler-Lagrange equation
1. If F = F (x, y 0 ), i.e., F does not explicitly depend on
y (x) the E-L equation reduces to
d ∂F ∂F
0
dx = 0 ⇒ = const. (9)
dx ∂y ∂y 0
2. If F = F (y , y 0 ), i.e., F is independent of x then the
E-L lagrange equation becomes
∂F
F − y0 = const (10)
∂y 0
In both cases we are left with a first- instead of a second-order
ODE.
The Brachistochrone problem
Perhaps the first problem in the calculus of variations was the
brachistochrone problem formulated by J. Bernoulli in 1696:
Consider a bead sliding under gravity along a smooth
wire joining two fixed points A and B (not on the same
vertical line).
What is the shape of the wire in order that the bead,
when released from rest at point A, slides to B in
minimum time?
A a
b
B
The Brachistochrone problem
The starting point A = (0, 0) , the end point B(a, b) .
The shape of the wire y = y (x), y axis points downwards.
1
The kinetic energy of the bead is mv 2 , m is the mass.
2
Due to our sign convention, the potential energy at height
y = y (x) is −mgy (x) , g the gravitational acceleration.
The bead is initially at rest,with 0 kinetic energy and 0
potential energy. Assuming that frictional forces are
negligible, conservation of energy implies that the total
1 2
energy must remain equal to 0: mv − mgy = 0 .
2
A a
b
B
The Brachistochrone problem
Using the energy conservation law we find
s
1 + y 02
Z Z
1 ds 1
t=√ √ =√ dx (11)
2g y 2g y
The problem is now to minimise
s
1 + y 02
Z
1
J[y ] = √ dx (12)
2g y
We solve the E-L equation. The solutions is vertically inverted
cycloid
x = A (φ − sin φ) (13)
y = A (1 − cos φ)
Substituting the solution (13) into (12) and integrating
p √
t̄ = 2g t = 2Aφ
The Brachistochrone problem
p √
t̄ = 2g t = 2Aφ
2 4 6 8 10 12
A=1
2
K P1 Q
A=2
P2
Fig.: Cycloid for A = 1 and A = 2. A cycloid is the curve
traced by a point on the rim of a circular wheel with radius A
as the wheel rolls along a straight line without slipping.
Isoperimetric Problems
The variational problem with integral type constraint is to find
a function y (x) for which the functional
Z b
Φ[y ] = F (x, y , y 0 )dx (14)
a
has an extremum and for which the constraint equation
Z b
G (x, y , y 0 )dx = C , C ∈R (15)
a
and the boundary conditions
y (a) = y1 and y (b) = y2
are satisfied.
The simplest isoperimetric problem
Given a rope of fixed length and a line segment (the
x-axis), what shape should the rope take so that the area
enclosed by the rope and the line segment is maximised?
Thus it is required to find a curve y(x) ≥ 0 of a fixed
length l that encloses the largest area with the x axis, i.e
maximize the functional
Z b Z bq
I(y) = y(x)dx for J[y] = 1 + y0 2 dx = l. (16)
a a
Isoperimetric Problems
Solve the Euler-Lagrange equation
∂H d ∂H
− = 0 with H = F + λG. (17)
∂y dx ∂y0
The new variable λ is called a Lagrange multiplier.
After finding a general solution of (17) the constants in
the solution and the Langrange multiplier can be
determined from the boundary conditions
y(a) = y1 and y(b) = y2
and the constrained equation
Z b
G(x, y, y0 )dx = C. (18)
a
.
The hanging chain
Consider a flexible homogeneous thin chain, rope or cable of
length L hanging between the two points (x1 , y1 ) and (x2 , y2 ).
We want to determine the equation of the curve formed
by the chain supported at its ends.
We assume that the cable is in a stable equilibrium
configuration so that the potential energy is a minimum.
At a general point (x, y ) on the hanging chain consider
an element of mass dm. It has the potential energy
dV = (dm)gy = (ρds)gy,
where g is the gravitational
p acceleration, ρ is the mass
0 2
density and ds = 1 + (y ) dx is an element of arc
length along the cable.
The hanging chain
The total potential energy of the cable is obtained by
integration over the length of the cable
Z Z q
V = dV = ρg y 1 + (y 0 )2 dx. (19)
The problem is to find the curve y = y (x) such that the
integral given by equation (19) is a minimum subject to the
constraint conditions
Z q
J= 1 + (y 0 )2 = L (20)
and the boundary conditions y (x1 ) = y1 and y (x2 ) = y2 .
Catenery
x − c2
y (x) = c1 cosh − λ.
c1
Fig.: Catenary is a curve that describes the shape of a flexible
hanging chain or cable.
The name derives from the Latin catenaria (chain). Any freely
hanging cable or string assumes this shape, if the body is of
uniform mass per unit of length and is acted upon solely by gravity.
Additional Variational Concepts
Variational problems considered so far:
Find y = y (x) ∈ C 1 [a, b] that extremises the functional
Z b
J[y ] = F (x, y , y 0 )dx, (21)
a
and satisfies Dirichlet or essential boundary conditions
y (a) = α, y (b) = β .
What if y (a) and y (b) are not given?
Additional Variational Concepts
The variational problem is now:
Find a function y = y (x) that minimizes
Z x2
J[y ] = L(x, y , y 0 )dx.
x1
but one or both of the end point conditions y (x1 ) = y1
and y (x2 ) = y2 are not prescribed.
Additional Variational Concepts
We considered
Z x2
h(ε) = J[y + εv ] = L(x, y + εv , y 0 + εv 0 )dx.
x1
For the stationary value y (x)
Z x2
dh ∂L ∂L
|ε=0 = v (x, y , y 0 ) + v 0 0 (x, y , y 0 ) dx = 0.
dε x1 ∂y ∂y
We integrated the second term using integration by parts
Additional Variational Concepts
Z x2
dh ∂L x2 ∂L 0 d ∂L 0
|ε=0 = v |x1 + v (x, y , y ) − (x, y , y ) dx = 0.
dε ∂y 0 ∂y dx ∂y 0
x1
| {z }
Euler-Lagrange equation, =0
∂L x2 ∂L ∂L
0
v |x1 = 0
|x=x2 v (x2 ) − 0 |x=x1 v (x1 ) = 0.
∂y ∂y ∂y
In the case of y (x1 ) = y1 , y (x2 ) = y2 , we required that
v (x1 ) = 0 and v (x2 ) = 0.
In the case y (x1 ), y (x2 ) are not given, one must require
∂L ∂L
|x=x1 = 0 and |x=x2 = 0 (22)
∂y 0 ∂y 0
Natural or Neumann Boundary conditions
The Brachistochrone problem
A a
b
B
s
1 + y 02
Z
1
J[y ] = √ F (x, y , y 0 )dx, F = , y (0) = 0, y (a) = b.
2g y
(23)
The solutions of E-L equation for F (x, y , y 0 )
x = A (φ − sin φ) (24)
y = A (1 − cos φ)
Now we consider the shortest time, with y (0) = 0 and
∂F
|x=b = 0. (25)
∂y 0
Various Generalizations
For two functions, the functional is
Z b
I [y , z] = F (x, y , y 0 , z, z 0 ) dx .
a
Stationary points, subject to y (a), y (b), z(a) and z(b) being fixed
require
∂F d ∂F ∂F d ∂F
− =0 , − =0 . (26)
∂y dx ∂y 0 ∂z dx ∂z 0
Here the Euler-Lagrange equations form a system of two coupled
second-order ODEs for y (x) and z(x).
Various Generalizations
Higher derivatives of the argument function
Z b
I [y ] = F (x, y , y 0 , y 00 ) dx .
a
The E-L equation for given y (a), y (b), y 0 (a) and y 0 (b) is given by
d2
∂F d ∂F ∂F
− + 2 =0, (27)
∂y dx ∂y 0 dx ∂y 00
is a fourth-order ODE.
Various Generalizations
Several independent variables
Z
∂φ
I [φ] = F (x, y , φ, φx , φy )dx dy with φx ≡ etc.
C1 ∂x
If φ is fixed everywhere on the boundary of the integration region
C1 , then the E-L equation is the second-order partial
differential equation (PDE)
∂F d ∂F d ∂F
− − =0. (28)
∂φ dx ∂φx dy ∂φy
Informal Module Questionnaire
An informal module questionnaire has been placed on
Canvas under Quizzes. Please click on Quizzes then on
the right panel ”Informal module questionnaire” then
”Continue”.
The questionnaire is anonymous, and the purpose is to
allow you to express your opinion on how the module
has been delivered so far and what can be changed to
make the delivery of the module better for you.