Calculus of Variations

Ramesh Kadambi
January 18, 2025

1
May, 09, 2021

Contents
1 Introduction 3

2 Function Spaces 3

3 The Variation and the Necesssry condition for an Extremum 4

4 Minima and Maxima of functions of several variables 5

4.1 The function space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4.2 Some useful Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

5 Euler’s Equation 6
May, 09, 2021

1 Introduction
Calculus of variations is the methodology used to minimize or minimize functionals.

Definition 1 (Functional). A functional is a mapping f : F 7→ R where F belonging to some class of functions.

In simple terms a functional is a function whose independent variable itself is a function.

Rb
Example 1. A standard example is J[y] = a F [x, y(x), y ′ (x)]dx where y(x) ranges over set of all continuously differential
functions.
Rb
Definition 2 (Localization Property). A functional a F (x, y, y ′ )dx is said to have localization property if for every partition of
Rb Pn R x
[a, b] we have a F (x, y, y ′ )dx = i=0 xii+1 F (x, y, y ′ )dx.

The linking to the standard analysis uses this localization property. Given a partition as before,
Z b n Z xi+1 n
X X yi − yi−1
F (x, y, y ′ )dx = F (x, y, y ′ )dx = F (x, y, )h
a i=0 xi i=0
h

where y ′ = yi −y
h
i−1
and h = yi −yi−1 . This reduces the integral to a function of n variables and the problem looks like multivariate
optimization. Limits as n → ∞ will produce exact solution. Functions therefore can be regarded as functions of infinitely many
variables. This reduces the problems of calculus of variations to an analog of differential calculus.

2 Function Spaces
Function spaces are sets where the elements are functions.

Definition 3 (Linear Space). A linear space is a set R of elements of any kind for which the operation of addition and scalar
multiplication by real numbers are defined and follow the rules,
1. ∀ x, y ∈ R we have x + y = y + x ∈ R.

2. ∀α ∈ R and ∀x ∈ R we have αx ∈ R.
3. There exists a 0 ∈ R such that ∀x ∈ R we have x + 0 = x.
4. ∀x ∈ R we have an element −x ∈ R such that x + (−x) = 0.
5. ∀x ∈ R we have 1.x = x.

6. ∀α, β ∈ R and x ∈ R we have α(βx) = (αβ)x, (α + β)x = αx + βx, and α(x + y) = αx + αy.

Definition 4 (Normed Space). A linear space R is said to be normed, if each element x ∈ R is assigned a non-negative number
∥x∥ such that,

1. ∥x∥ ≥ 0.
2. ∀α ∈ R and ∀x ∈ R we have ∥αx∥ = |α|∥x∥.
3. ∀x, y ∈ R we have ∥x + y∥ ≤ ∥x∥ + ∥y∥.

Definition 5 (Continuity of Functional). The functional J[y] is said to be continuous at the point ŷ ∈ R if for any ϵ > 0, there
is a δ > 0 such that

|J[y] − J[ŷ]| < ϵ (2.1)

provided that ∥y − ŷ∥ < δ.

Definition 6 (Lower and Upper Semi-Continuous). If in the definition of continuity we replace the condition (2.1) by J[y]−J[ŷ] >
−ϵ is lower semi-continuous at ŷ, and J[y] − J[ŷ] < ϵ is upper semi-continuous at ŷ.
May, 09, 2021

The question becomes under what space and metric should we study these functionals? The space of all continuous functions
Rb
is insufficient. The choice of the space is mostly based on the spaces where the functionals of the form J[y] = a F (x, y, y ′ )dx.
This functional is not necessarily continuous on the space of continuous functions, it is however continuous in the space of
functions that are at least once differentiable with the norm,

∥y∥ = max |y(x)| + max |y ′ (x)|.

a≤x≤b a≤x≤b

In many cases the solutions to these variational problems may not form a normed linear vector space. In those cases we
restrict to admissible set of functions.

3 The Variation and the Necessary condition for an Extremum

Definition 7 (Continuous Functional). Given a normed linear space R, let each element h ∈ R be assigned a number ϕ[h], i.e.,
let ϕ[h] be a functional defined on R, i.e. ϕ : R 7→ R. Then ϕ[h] is said to be a continuous linear functional if
1. ϕ[αh] = αϕ[h] ∀h ∈ R,
2. ϕ[h1 + h2 ] = ϕ[h1 ] + ϕ[h2 ] ∀h1 , h2 ∈ R,
3. ϕ[h] is continuous ∀h ∈ R.

A continuous linear functional is continuous mapping that preserves a linear structure.

Lemma 1. If α(x) is continuous in [a, b], and if

Z b
α(x)h(x)dx = 0
a

for every function h(x) ∈ C(a, b) such that h(a) = 0, h(b) = 0, then α(x) = 0 for all x ∈ [a, b].
Proof. Let there be a α(x) such that α(x) > 0, x ∈ [c, d] and for every h(x) ∈ C. Let us define h(x) = (x − c)(d − x) on the
interval [c, d] and zero elsewhere. Now clearly h(x) ≥ 0 and,
Z d Z b
α(x)h(x) = α(x)(x − c)(d − x)dx > 0
c a

This is a contradiction. The trick here is to define a function only on a region that has a non-zero positive value or negative
value. The condition that zero condition holds for all functions h(x) ∈ C(a, b) is crucial. It should also be noted that the lemma
holds for functions in Dn (a, b).

Definition 8 (Differential of a Functional). Given J[y] defined on some normed linear space. We have ∆J[y] = J(y + h] − J[y]
as the increment, corresponding to the increment h = h(x) of the independent variable y = y(x). Fixing y we see that ∆J is a
functional in h. In general this is a non-linear functional. If we now have,

∆J = ϕ[h] + ϵ∥h∥

where ϕ[h] is a linear functional and ϵ → 0 as ∥h∥ → 0. Then the functional J[y] is said to be differentiable, and the principal
linear part of the increment ∆J[h], i.e., the linear functional ϕ[h] which differs from ∆J[h] by an infinitesimal of order higher
than 1 relative to ∥h∥, is called the variation (or differential) of J[y] and is denoted by δJ[h] = ϕ(h).

ϕ[h]
Lemma 2 (Linear Functional Limits). If ϕ[h] is a linear functional such that lim∥h∥→0 ∥h∥ = 0∀h ∈ R, then ϕ[h] = 0 for all
h, this is denoted by ϕ[h] ≡ 0.
ϕ[h0 ]
Proof. Let us assume for a minute that there exists a function h0 such that ϕ[ h0 ] ̸= 0 and h0 = λ. Consider the sequence of
functions h0
it is clear that ∥hn ∥ → 0 as n → ∞. Consider
n , limn→∞ ϕ[h
hn
n]
= limn→∞ nh0 = limn→∞ nϕ[h
ϕ[nh0 ]
nh0
0]
= λ. This is a
contradiction.
May, 09, 2021

Theorem 1 (Uniqueness of differentiable functional). The differential of a differentiable functional is unique.

Proof. We need to show that the linear functional ϕ(h) is unique. Since the functional is differentiable, we have,

∆J = J[y + h] − J[y] = ϕ[h] + ϵ∥h∥

ϕ[h] = ∆J − ϵ∥h∥

If there are two linear functionals then we have,

ϕ1 [h] = ∆J − ϵ1 ∥h∥
ϕ2 [h] = ∆J − ϵ2 ∥h∥

Subtracting the two we have the following,

ϕ[h] = ϕ1 [h] − ϕ2 [h] = (ϵ1 − ϵ2 )∥h∥

ϕ1 [h]−ϕ2 [h]
Taking limit as the ∥h∥ → 0 we have ∥h∥ = 0, it can be shown from here that the above implies that ϕ1 [h] − ϕ2 [h] = 0.

4 Minima and Maxima of functions of several variables

In the case of function of several variables F (x1 , x2 , · · · , xn ) is said to have a relative extremum (relative) at the point
(x1 , x2 , · · · , xn ) if ∆F = F (x1 , x2 , · · · , xn ) − F (x̂1 , x̂2 , · · · , x̂n ), has the same sign for all pints (x1 , x2 , · · · , xn ) belonging to
some neighborhood Br (x̂1 , x̂2 , · · · , x̂n ) where the extremum is a minimum ∆F ≥ 0 and a maximum ∆F ≤ 0.

The same idea is extended to the space of functions.

Definition 9 (Extremum of Functionals). A functional J[y] has a (relative) extremum for y = ŷ if J[y] − J[ŷ] does not change
its sign in some neighborhood of the function (curve) y = ŷ. As in the case of functions of several variable ∆J = J[y] − J[ŷ] > 0
at a minima and ∆J = J[y] − J[ŷ] < 0 at a maxima.

4.1 The function space

The question now becomes of what space should one use to prove existence and find these extrema. These spaces are defined by
an appropriate norm defined on them.
Example 2 (Space Continuous Functions). The space of continuous funtions C(a, b) with the norm ∥y∥0 = maxa≤x≤b |y(x)| is
a normed vector space.
Example 3 (The space of differentiable functions). The space D1 (a, b) with norm ∥y∥1 = maxa≤x≤y |y(x)| + maxa≤x≤b |y ′ (x)|.
The typical functional considered is continuous in the space of once differentiable functions D1 and not so much in the space
of continuous functions C.
Definition 10 (Weak Extrema). The functional J[y] has a weak extremum for y = ŷ if there exists an ϵ > 0 such that J[y] − J[ŷ]
has the same sign for all y in the domain of definition of the functional which satisfy the condition ∥y − ŷ∥1 < ϵ. Where ∥.∥1 is
norm on the space D1 .

Definition 11 (Strong Extrema). The functional J[y] has a strong extremum for y = ŷ if there exists an ϵ > 0 such that
J[y] − J[ŷ] has the same sign for all y in the domain of definition of the functional which satisfy the condition ∥y − ŷ∥0 < ϵ.
Where ∥.∥0 is norm on the space C.

The above definitions can be formulated in terms of open sets or open balls. Where the norm is ∥y = ŷ∥ replaced by Bϵ (ŷ).

Lemma 3 (Strong vs Weak Extremum). Strong extremum implies weak extremum, Extremum(∥.∥0 ) ⇒ Extremum(∥.∥1 ).
Proof. This is fairly straight forward to show, let ŷ be an extremum under ∥ŷ∥0 and J[y] − J[ŷ] > 0 for some Bϵ0 (ŷ), i.e. ŷ is an
upper bound on all the feasible set of functions, adding differentiability will not result in a more optimal solution. On the other
hand if ŷ is an Extremum(∥.∥1 ), then there exists Bϵ1 (ŷ) such that ∆J > 0, however this does not imply there exists Bϵ0 (ŷ)
where ∆J > 0 once you add all the continuous functions back. This proof is a bit odd, essentially the weaker form optimum is
the upper/lower bound of the solutions possible.
May, 09, 2021

The bottom line is that the stronger condition allows for larger set and possibility of a higher optimum value.
Theorem 2 (Necessary Condition for Extremum). A necessary condition for the differentiable functional J[y] to have an
extremum for y = ŷ is that its variation vanish for y = ŷ, i.e. that

δJ[h] = 0

, for y = ŷ and all admissible h.

Since the functional is differentiable we can write the differential as follows,
∆J[y] = J[y] − J[ŷ] = δJ[h] + ϵ∥h∥
For sufficiently small ∥h∥ since ϵ → 0 as ∥h∥ → 0. The bottom line is that sign of ∆J[h] is the same as the sign of δJ[h].
Consider an h0 such that δJ[h0 ] ̸= 0, for some admissible h0 . For some α > 0, δJ[−αh0 ] = −αδJ[h0 ]. Now if J[y] has a
minimum at ŷ then the fact that we can make the sign of −αδJ[h] anything we want creates a contradiction. This implies that
∆J[h] = J[ŷ + h] − J[ŷ] ≥ 0.

4.2 Some useful Lemmas

Lemma 4. If α(x) is continuous in (a, b), and if
Z b
α(x)h′ (x)dx = 0
a

for every h(x) ∈ D(a, b) such that h(x) = h(b) = 0, then α(x) = c for all x ∈ [a, b] where c is a constant.

Proof. Just integrating by parts we have,

Z b Z b
α(x)h′ (x)dx = − α′ (x)h(x)dx = 0
a a

Now we use our previous lemma that −α′ (x) = 0 integrating we have α(x) = c.

Lemma 5. If α(x) is continuous in (a, b), and if

Z b
α(x)h′′ (x)dx = 0
a

for all h(x) ∈ D2 (a, b) such that h(a) = h(b) = 0 and h′ (a) = h′ (b) = 0, then α(x) = c0 + c1 x for all x ∈ [a, b], where c0 , c1 are
constants.

Proof. The proof is as before, integrating by parts twice we have,

Z b Z b
′′
α(x)h (x)dx = α′′ h(x) = 0
a a

Using our earlier lemma, we obtain the following.

α′′ = 0 ⇒ α = c0 + c1 x

5 Euler’s Equation
We now look at the most simple of problems in variational calculus.
Problem 1 (Euler’s Problem). Let F (x, y, z) be a function that contains first and second partial derivatives with respect to all
its arguments. Then among all functions y(x) which are continuously differentiable for a ≤ x ≤ b and satisfy the boundary
conditions
y(a) = A, y(b) = B
May, 09, 2021

find the function for which the functional,

Z b
J(y) = F (x, y, y ′ )dx
a

has a weak extremum.

Solution 1. This is pretty straight forward. We consider the variation of the functional and then use Taylor expansion.

Z b
∆J = J[y + h] − J[y] = (F (x, y + h, y ′ + h′ )dx − F (x, y, h)) dx
a

Using Taylor expansion we have the following,

Z b
∆J = (Fy h + Fy′ h′ ) dx + · · ·
a

Using our earlier lemma we obtain the following,

d
Fy − Fy ′ = 0
dx