8

CALCULUS OF VARIATIONS

INTRODUCTION

The calculus of variations and its extensions are devoted to finding the
optimum function that gives the best value of the economic model and
constraints of a system. The need for an optimum function, rather than an
optimal point, arises in numerous problems from a wide range of fields in
engineering and physics, which include optimal control, transport
phenomena, optics, elasticity, vibrations, statics and dynamics of solid
bodies, and navigation. Two examples are determining the optimal
temperatures profile in a catalytic reactor to maximize the conversion and
the optimal trajectory for a missile to maximize the satellite payload placed
in orbit. The first calculus of variations problem, the Brachistochrone
problem, was posed and solved by Johannes Bernoulli in 1696 (1). In this
problem the optimum curve was determined to minimize the time traveled
by a particle sliding without friction between two points.
This chapter is devoted to a relatively brief discussion of some of the key
concepts of this topic. These include the Euler equation and the Euler-
Poisson equations for the case of several functions and several independent
variables with and without constraints. It begins with a derivation of the
Euler equation and extends these concepts to more detailed cases. Examples
are given to illustrate this theory.
The purpose of this chapter is to develop an appreciation for what is
required to determine the optimum function for a variational problem. The
extensions and applications to optimal control, Pontryagin’s maximum
principle, and continuous dynamic programming are left to books devoted to
those topics.

FUNCTIONS, FUNCTIONALS, AND NEIGHBORHOODS

It will be necessary to discuss briefly functionals and neighborhoods before

developing the Euler equation for the solution of the simplest problem in the
calculus of variations. In mathematical programming the maximum or
minimum of a function was determined to be an optimal point or set of
points. In the calculus of variations, the maximum or minimum value of a
functional is determined to be an optimal function. A functional is a function
of a function and depends on the entire path of one or more functions, rather
than a number of discrete variables.
 For the calculus of variations the functional is an integral, and the function
that appears in the integrand of the integral is to be selected to maximize or
minimize the value of the integral. The texts by Forray (1), Ewing (2),
Weinstock (3), Schechter (4), and Sagan (6) elaborate on this concept.
However, at this point let us examine an example of the functional given by
equation (8-1). The minimum of this functional is a function y(x) that gives
the shortest distance between two points [𝑥0 , 𝑦(𝑥0 )] and [𝑥1 , 𝑦(𝑥1 )].

𝒙𝟏
𝑰[𝒚(𝒙)] = ∫ [𝟏 + (𝒚′ )𝟐 ]𝟏⁄𝟐 (8-1)
𝒙𝟎

In this equation 𝑦′ is the first derivative of 𝑦 with respect to 𝑥. The function

that minimizes this integral, a straight line, will be obtained as an illustration
of the use of the Euler equation in the next section.
The concept of a neighborhood is used in the derivation of the Euler
equation to convert the problem into one of finding the stationary point of a
function of a single variable. A function 𝑦̅ is said to be in the neighborhood
of a function 𝑦 if |𝑦̅ − 𝑦| ≤ ℎ (5). This is illustrated in Figure 8-1(a). The
concept can be extended for more restrictive conditions, such as that shown
in Figure 8-1(b), when |𝑦̅ − 𝑦| ≤ ℎ and |𝑦̅ ′ − 𝑦 ′ | ≤ ℎ. For this case 𝑦̅ is said
to be in the neighborhood of first order to 𝑦. Consequently, the higher the
order of the neighborhood, the more nearly the functions will coincide.
Extensions of these definitions will lead to what are referred to as strong and
weak variations (6).

EULER EQUATION

The simplest form of the integral to be optimized by the calculus of variations

is the following one:
𝒙𝟏
𝑰[𝒚(𝒙)] = ∫ 𝑭(𝒙, 𝒚, 𝒚′ ) 𝒅𝒙 (8-2)
𝒙𝟎

In addition, the values of 𝑦(𝑥0 ) and 𝑦(𝑥1 ) are known, and an example of the
function 𝐹(𝑥, 𝑦, 𝑦 ′ ) was given in equation (8-1) as:

𝑭(𝒙, 𝒚, 𝒚´ ) = [𝟏 + (𝒚′ )𝟐 ]𝟏⁄𝟐 (8-3)

To obtain the optimal function that minimizes the equation (8-2), it is

 (𝑎) |𝑦̅ − 𝑦| ≤ ℎ

(𝑏) |𝑦̅ − 𝑦| ≤ ℎ 𝑎𝑛𝑑 |𝑦̅ ′ − 𝑦 ′ | ≤ ℎ

Figure 8-1. Illustration of the concept of a neighborhood.

necessary to solve the Euler equation, which is the following second-order

ordinary differential equation.
𝒅 𝝏𝑭 𝝏𝑭
( ´) − =𝟎 (8-4)
𝒅𝒙 𝝏𝒚 𝝏𝒚

It is not obvious that equation (8-4) is a second-order ordinary differential

equation. Also, it probably appears unusual to be partially differentiating the
function 𝐹 with respect to 𝑦 and 𝑦 ′ . In addition, although the term minimize
will be used, stationary points are being located, and their character will have
to be determined using sufficient conditions. Consequently, it should be
beneficial to outline the derivation of the Euler equation.
First, 𝑦(𝑥) is specified as the function that minimizes the functional
𝐼[𝑦(𝑥)], equation (8-2). (However, the form of 𝑦(𝑥) has to be determined.)
Then a function 𝑦̅(𝑥) is constructed to be in the neighborhood of 𝑦(𝑥) as
follows:

̅(𝒙) = 𝒚(𝒙) + 𝜶𝒏(𝒙)

𝒚 (8-5)
where 𝛼 is a parameter that can be made arbitrarily small. Also 𝑛(𝑥) is a
continuously differentiable function defined on the interval 𝑥0 ≤ 𝑥 ≤ 𝑥1
with 𝑛(𝑥0 ) = 𝑛(𝑥1 ) = 0, but is arbitrary elsewhere. The results from the
derivation using equation (8-5) are described mathematically as weak
variations (1), for 𝛼𝑛(𝑥) and 𝛼𝑛(𝑥)′ are small.
Now equation (8-2) is written in terms of the function 𝑦̅(𝑥) as:

𝒙𝟏
̅(𝒙)] = ∫ 𝑭(𝒙, 𝒚
𝑰[𝒚 ̅′ )𝒅𝒙
̅, 𝒚 (8-6)
𝒙𝟎

The above equation can be put in terms of the optimal function 𝑦(𝑥) and the
arbitrary function 𝑛(𝑥) using equation (8-5).
𝒙𝟏
̅(𝒙)] = ∫ 𝑭(𝒙, 𝒚 + 𝜶𝒏, 𝒚′ + 𝜶𝒏′ )𝒅𝒙
𝑰[𝒚 (8-7)
𝒙𝟎

The mathematical argument (3) is made that all the possible functions 𝑦̅
lie in an arbitrarily small neighborhood of y, because 𝛼 can be made
arbitrarily small. As such, the integral of equation (8-7) may be regarded as
an ordinary function of 𝛼, Φ(𝛼), because 𝛼 would specify the value of the
integral knowing Φ(𝛼 = 0) at the minimum from 𝑦(𝑥).

𝒙𝟏
̅(𝒙)] = Φ(𝜶) = ∫ 𝑭(𝒙, 𝒚 + 𝜶𝒏, 𝒚′ + 𝜶𝒏′ )𝒅𝒙
𝑰[𝒚 (8-8)
𝒙𝟎

The minimum of Φ(𝛼) is obtained by sitting the first derivative of Φ with

respect to 𝛼 equal to zero. The differentiation is indicated as:

𝒅𝚽(𝜶) 𝒅 𝒙𝟏
= ∫ 𝑭(𝒙, 𝒚 ̅′ )𝒅𝒙
̅, 𝒚 (8-9)
𝒅𝜶 𝒅𝜶 𝒙𝟎

Leibnitz' rule, equation (8-10), is required to differentiate the integral given

in equation (8-9):

𝒅 𝜶𝟐 𝒅𝒇 𝒅𝜶𝟐 𝒅𝜶𝟏
∫ 𝒅𝒙 + 𝒇(𝜶𝟐 , 𝒕) − 𝒇(𝜶𝟏 , 𝒕) (8-10)
𝒅𝒕 𝜶𝟏 𝒅𝒕 𝒅𝒕 𝒅𝒕

where 𝑥0 and 𝑥1 correspond to 𝛼1 and 𝛼2 , and 𝛼 corresponds to 𝑡.

The upper and lower limits, 𝑥1 and 𝑥2 , are constants; and the following
derivatives in the second and third terms on the right-hand side of equation
(8-10) are zero for this case:

𝒅𝒙𝟎 𝒅𝒙𝟏
= =𝟎 (8-11)
𝒅𝜶 𝒅𝜶

Consequently, the order of integration and differentiation is interchanged,

and equation (8-9) can be written as:

𝒙𝟏
𝒅𝚽(𝜶) 𝒅
=∫ ̅′ )𝒅𝒙
̅, 𝒚
𝑭(𝒙, 𝒚 (8-12)
𝒅𝜶 𝒙𝟎 𝒅𝜶

The integrand can be expanded as follows:

̅ 𝝏𝑭 𝒅𝒚
𝒅𝑭 𝝏𝑭 𝒅𝒚 ̅′ 𝝏𝑭 𝒅𝒙
= + ′ + (8-13)
̅ 𝒅𝜶 𝝏𝒚
𝒅𝜶 𝝏𝒚 ̅ 𝒅𝜶 𝝏𝒙 𝒅𝜶

where 𝑑𝑥 ⁄𝑑𝛼 = 0, because 𝑥 is treated as a constant in the mathematical

argument of considering changes from curve to curve at constant 𝑥.
Substituting equation (8-13) into equation (8-12) gives:

𝒅𝜱 𝒙𝟏
̅ 𝝏𝑭 𝒅𝒚
𝝏𝑭 𝒅𝒚 ̅′
=∫ [ + ′ ] 𝒅𝒙 (8-14)
𝒅𝜶 𝒙𝟎 𝝏𝒚̅ 𝒅𝜶 𝝏𝒚̅ 𝒅𝜶

The following results are needed:

̅
𝒅𝒚 𝒅 ̅′
𝒅𝒚 𝒅 ′
= [𝒚 + 𝜶𝒏] = 𝒏 = [𝒚 + 𝜶𝒏′ ] = 𝒏′ (8-15)
𝒅𝜶 𝒅𝜶 𝒅𝜶 𝒅𝜶

Using equation (8-15), we can write equation (8-14) as:

𝒙𝟏
𝒅𝜱 𝝏𝑭 𝝏𝑭
= ∫ [ 𝒏 + ′ 𝒏′ ] 𝒅𝒙 (8-16)
𝒅𝜶 𝒙𝟎 𝝏𝒚̅ ̅
𝝏𝒚

An integration-by-parts will give a more convenient form for the term

involving 𝑛′ , i.e.:

𝒙𝟏 𝒙𝟏
𝝏𝑭 ′ 𝝏𝑭 𝒙 𝒅 𝝏𝑭 (8-17)
∫ ′
𝒏 𝒅𝒙 = ′ 𝒏|𝒙𝟏𝟎 − ∫ 𝒏 ( ′ ) 𝒅𝒙
𝒙𝟎 ̅
𝝏𝒚 ̅
𝝏𝒚 𝒙𝟎 ̅
𝒅𝒙 𝝏𝒚
The first term on the right-hand side is zero, for 𝑛(𝑥0 ) = 𝑛(𝑥1 ) = 0.
Combining the results from equation (8-17) with equation (8-16) gives:

𝒙𝟏
𝒅𝜱 𝝏𝑭 𝒅 𝝏𝑭
= ∫ 𝒏(𝒙) [ − ( ′ )] 𝒅𝒙
𝒅𝜶 𝒙𝟎 ̅ 𝒅𝒙 𝝏𝒚
𝒅𝒚 ̅

At the optimum 𝑑𝛷⁄𝑑𝛼 = 0, and letting 𝛼 → 0 has 𝑦̅ → 𝑦 and 𝑦̅ ′ → 𝑦 ′ .

Therefore, the above equation becomes:
𝒙𝟏
𝝏𝑭 𝒅 𝝏𝑭
∫ 𝒏(𝒙) [ − ( )] 𝒅𝒙 = 𝟎 (8-18)
𝒙𝟎 𝝏𝒚 𝒅𝒙 𝝏𝒚′

To obtain the Euler equation, the fundamental lemma of the calculus of

variation is used. This lemma can be stated, after Weinstock (3), as:

If 𝑥0 and 𝑥1 (> 𝑥0 ) are fixed constants and 𝐺(𝑥) is a particular continuous

function in the interval 𝑥0 ≤ 𝑥 ≤ 𝑥1 and if:

𝒙𝟏
∫ 𝒏(𝒙) 𝑮(𝒙)𝒅𝒙 = 𝟎
𝒙𝟎

for every choice of the continuously differentiable function 𝑛(𝑥) for which
𝑛(𝑥0 ) = 𝑛(𝑥1 ) = 0, then 𝐺(𝑥) = 0 identically in the interval 𝑥0 ≤ 𝑥 ≤ 𝑥1 .

The proof of this lemma is by contradiction and is given by Weinstock (3).

Applying this lemma to equation (8-18) gives the Euler equation:

𝝏𝑭 𝒅 𝝏𝑭
− ( )=𝟎 (8-4)
𝝏𝒚 𝒅𝒙 𝝏𝒚′

This equation is a second-order ordinary differential equation and has

boundary conditions 𝑦(𝑥0 ) and 𝑦(𝑥1 ). The solution of this differential
equation 𝑦(𝑥) optimizes the integral 𝐼[𝑦(𝑥)].
A more convenient form of the Euler equation can be obtained by
applying the chain rule to 𝜕𝐹(𝑥, 𝑦, 𝑦 ′ )⁄𝜕𝑦 ′ :

𝝏𝑭 𝝏 𝝏𝑭 𝝏 𝝏𝑭 𝝏 𝝏𝑭
𝒅 ( ′ ) = ′ ( ′ ) 𝒅𝒚′ + ( ′ ) 𝒅𝒚 + ( ) 𝒅𝒙 (8-19)
𝝏𝒚 𝝏𝒚 𝝏𝒚 𝝏𝒚 𝝏𝒚 𝝏𝒙 𝝏𝒚′
or

𝒅 𝝏𝑭 𝝏𝟐 𝑭 𝒅𝟐 𝒚 𝝏𝟐 𝑭 𝒅𝒚 𝝏𝟐 𝑭
( ′) = ′𝟐 𝟐 + + (8-20)
𝒅𝒙 𝝏𝒚 𝝏𝒚 𝒅𝒙 𝒅𝒚𝝏𝒚′ 𝒅𝒙 𝝏𝒙𝝏𝒚′

Substituting equation (8-20) into equation (8-4) and rearranging gives a more
familiar form for a second-order ordinary differential equation:

𝝏𝟐 𝑭 𝒅𝟐 𝒚 𝝏𝟐 𝑭 𝒅𝒚 𝝏𝟐 𝑭 𝝏𝑭
𝟐 𝒅𝒙𝟐
+ ′ 𝒅𝒙
+ ′
− =𝟎 (8-21)
𝝏𝒚 ′ 𝝏𝒚𝝏𝒚 𝝏𝒙𝝏𝒚 𝝏𝒚

A more convenient way to write this equation is:

𝒅𝟐 𝒚 𝒅𝒚
𝑭𝒚′ 𝒚′ + 𝑭𝒚′ 𝒚 + 𝑭𝒚′𝒙 − 𝑭𝒚 = 𝟎 (8-22)
𝒅𝒙𝟐 𝒅𝒙

The coefficients for the differential equation come from partially

differentiating 𝐹.
A special case that sometimes occurs is to have 𝐹(𝑦, 𝑦 ′ ), i.e., 𝐹 is not a
function of 𝑥. For this situation it can be shown that:

𝒅 𝝏𝑭
(𝑭 − 𝒚′ ′ ) = 𝟎 (8-23)
𝒅𝒙 𝝏𝒚

This equation may be integrated once to obtain a form of the Euler equation
given below, which can be a more convenient starting point for problem
solving:

𝝏𝑭
𝑭 − 𝒚′ = 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕 (8-24)
𝝏𝒚′

where the constant is evaluated using one of the boundary conditions.

At this point it should be noted that the necessary conditions of the
classical theory of maxima and minima have been used to locate a stationary
point. This point may be a minimum, maximum, or saddle point. To
determine its character, sufficient conditions must be used, and these will be
discussed subsequently. However, before that the following example is used
to illustrate an application of the Euler equation.
EXAMPLE 8-1

Determine the function that gives the shortest distance between two given
points. Referring to Figure 8-2, we can state the problem as:
𝒙𝟏
𝑴𝒊𝒏𝒊𝒎𝒊𝒛𝒆: 𝑳 = ∫ 𝒅𝒔
𝒙𝟎

and from the figure it follows that:

𝒅𝒔 = [(𝒅𝒙)𝟐 + (𝒅𝒚)𝟐 ]𝟏⁄𝟐 = [𝟏 + (𝒚′ )𝟐 ]𝟏⁄𝟐 𝒅𝒙

Substituting for 𝑑𝑠 in the integral gives:

𝒙𝟏
𝑳 = ∫ [𝟏 + (𝒚′ )𝟐 ]𝟏⁄𝟐 𝒅𝒙
𝒙𝟎

Figure 8-2. Diagram to illustrate the shortest distance between two points
for Example 8-1.
Evaluating the partial derivatives for the Euler equation:

𝑭𝒚 = 𝟎 𝑭𝒚′ 𝒙 = 𝟎 𝑭𝒚′ 𝒚 = 𝟎 𝑭𝒚′ 𝒚′ = 𝟏⁄[𝟏 + (𝒚′ )𝟐 ]𝟑⁄𝟐

Substituting, equation 8-22 becomes:

𝒅𝟐 𝒚 𝒅𝒚
(𝟏⁄[𝟏 + (𝒚′ )𝟐 ]𝟑⁄𝟐 ) 𝟐
+ (𝟎) + (𝟎) − (𝟎) = 𝟎
𝒅𝒙 𝒅𝒙

Simplifying gives

𝒅𝟐 𝒚
=𝟎
𝒅𝒙𝟐

Integrating the above equation twice gives:

𝒚 = 𝒄𝟏 𝒙 + 𝒄𝟐

This is the equation of a straight line, and the constants, 𝑐1 and 𝑐2 are
evaluated from the boundary conditions.

Another classic problem of the calculus of variation, as mentioned earlier,

is the Brachistochrone problem (10). The shape of the curve between two
points is to be determined to minimize the time of a particle sliding along a
wire without frictional resistance. The particle is acted upon only by
gravitational forces as it travels between the two points. The approach to the
solution is the same as for Example 8-1, and the integral for the time of travel,
𝑇, is given by Weinstock (3) as:

𝒙𝟏
𝑻 = ∫ {[𝟏 + (𝒚′ )𝟐 ]𝟏⁄𝟐 ⁄[𝟐𝒈(𝒚 − 𝒚𝟎 )]𝟏⁄𝟐 } 𝒅𝒙 (8-25)
𝒙𝟎

and the solution is in terms of the following parametric equations:

𝒙 = 𝒙𝟎 + 𝒂[𝜽 − 𝐬𝐢𝐧(𝜽)] 𝒚 = 𝒚𝟎 + 𝒂[𝟏 − 𝐜𝐨𝐬(𝜽)] (8-26)

The details of the solution are given by Weinstock (3). The solution is the
equations for a cycloid.
The method of obtaining the Euler equation is used almost directly to
obtain the results for more detailed forms of the integrand of equation (8-2).
The next section extends the results for more complex problems.
MORE COMPLEX PROBLEMS

In the procedure used to obtain the Euler equation, first a function was
constructed to have the integral be a function of a single independent
variable, 𝛼𝑖 , and then the classical theory of maxima and minima was applied
to locate the stationary point. This same method is used for more complex
problems that include more functions, e.g., 𝑦1 , 𝑦2 , …, 𝑦𝑛 ; in higher order
derivatives, e.g., 𝑦, 𝑦 ′ , …, 𝑦 (𝑛) ; and more than one independent variable,
e.g., 𝑦(𝑥1 , 𝑥2 ). It is instructive to take these additional complications in steps.
First, the case will be considered for one function 𝑦 with higher order
derivatives, and then this will be followed by the case of several functions
with first derivatives, all for one independent variable. These results can then
be combined for the case of several functions with higher order derivatives.
The results will be a set of ordinary differential equations to be solved. Then
further elaboration on the same ideas for the case of more than one
independent variable will give a partial differential equation to be solved for
the optimal function. Finally, any number of functions of varying order of
derivatives with several independent variables will require that a set of partial
differential equations be solved for the optimal functions.

Functional with Higher Derivatives in the Integrand: For the case of the
integrand containing higher order derivatives, the integral has the following
form:
𝒙𝟏
𝑰[𝒚(𝒙)] = ∫ 𝑭[𝒙, 𝒚, 𝒚′ , … , 𝒚(𝒎) ] 𝒅𝒙 (8-26)
𝒙𝟎

In this case, boundary conditions will be required for 𝑦(𝑥0 ), 𝑦 ′ (𝑥0 ),…,
𝑦 (𝑚) (𝑥0 ), and 𝑦(𝑥1 ), 𝑦 ′ (𝑥1 ),…, 𝑦 (𝑚) (𝑥1 ).
The function constructed in equation (8-5) is used, and the integral of
equation (8-26) becomes:
𝒙𝟏
̅(𝒙)] = 𝜱(𝜶) = ∫ 𝑭[𝒙, 𝒚
𝑰[𝒚 ̅′ , 𝒚
̅, 𝒚 ̅(𝒎) ] 𝒅𝒙
̅′′ , … , 𝒚 (8-27)
𝒙𝟎

The mathematical argument is used that the integral is a function of 𝛼 only,

and differentiation with respect to 𝛼 gives:

𝒙𝟏
𝒅𝜱 𝒅
=∫ ̅ , 𝒚̅ ′ , 𝒚̅ ′′ , … , 𝒚̅ (𝒎) ] 𝒅𝒙
𝑭[𝒙, 𝒚 (8-28)
𝒅𝜶 𝒙𝟎 𝒅𝜶
and using the chain rule, we can write the integrand as:

̅ 𝝏𝑭 𝒅𝒚
𝒅𝑭 𝝏𝑭 𝒅𝒚 ̅′ 𝝏𝑭 𝒅𝒚
̅′′ ̅(𝒎)
𝝏𝑭 𝒅𝒚
= + ′ + ′′ + ⋯ + (𝒎) (8-29)
̅ 𝒅𝜶 𝝏𝒚
𝒅𝜶 𝝏𝒚 ̅ 𝒅𝜶 𝝏𝒚̅ 𝒅𝜶 ̅
𝝏𝒚 𝒅𝜶

Using the function 𝑦̅ = 𝑦 + 𝛼𝑛 and its derivatives gives:

𝒅𝑭 𝝏𝑭 𝝏𝑭 𝝏𝑭 𝝏𝑭
= 𝒏 + ′ 𝒏′ + ′′ 𝒏′′ + ⋯ + (𝒎) 𝒏(𝒎) (8-30)
̅
𝒅𝜶 𝝏𝒚 ̅
𝝏𝒚 ̅
𝝏𝒚 ̅
𝝏𝒚

and equation (8-28) can be written as the following:

𝒙𝟏
𝒅𝜱
= ∫ [𝑭𝒚̅ 𝒏 + 𝑭𝒚̅′ 𝒏′ + 𝑭𝒚̅′′ 𝒏′′ + ⋯ + 𝑭𝒚̅(𝒎) 𝒏(𝒎) ] 𝒅𝒙 (8-31)
𝒅𝜶 𝒙𝟎

A series of integration-by-parts converts the terms in equation (8-31) as

follows:

𝒙𝟏 𝒙𝟏
𝒅
∫ 𝑭𝒚̅′ 𝒏′ 𝒅𝒙 = − ∫ 𝒏 𝑭 ′ 𝒅𝒙 (8-32)
𝒙𝟎 𝒙𝟎 𝒅𝒙 𝒚̅

𝒙𝟏 𝒙𝟏
𝒅 𝒙𝟏
𝒅𝟐
∫ 𝑭𝒚̅′′ 𝒏′′ 𝒅𝒙 = − ∫ 𝒏′ 𝑭𝒚̅′′ 𝒅𝒙 = ∫ 𝒏 𝟐 𝑭𝒚̅′′ 𝒅𝒙
𝒙𝟎 𝒙𝟎 𝒅𝒙 𝒙𝟎 𝒅𝒙
⋮
𝒙𝟏 𝒙𝟏
𝒅(𝒎)
∫ 𝑭𝒚̅(𝒎) 𝒏(𝒎) 𝒅𝒙 = (−𝟏)𝒎 ∫ 𝒏 (𝒎) 𝑭𝒚̅(𝒎) 𝒅𝒙 (8-33)
𝒙𝟎 𝒙𝟎 𝒅𝒙

where equation (8-32) is the same as equation (8-17).

Equation (8-31) can be written as:

𝒅𝜱 𝒙𝟏 𝒅𝑭𝒚̅ 𝒅(𝒎)
= ∫ 𝒏 [𝑭𝒚̅ − + ⋯ + (−𝟏)𝒎 (𝒎) 𝑭𝒚̅(𝒎) ] 𝒅𝒙 (8-34)
𝒅𝜶 𝒙𝟎 𝒅𝒙 𝒅𝒙

At the optimum 𝛼 → 0 to have 𝑦̅ → 𝑦, …, 𝑦̅ (𝑚) → 𝑦 (𝑚) ; and 𝑑𝛷⁄𝑑𝛼 = 0 to

give:

𝒅𝑭𝒚 𝒅(𝒎)
𝑭𝒚 − + ⋯ + (−𝟏)𝒎 (𝒎) 𝑭𝒚(𝒎) = 𝟎 (8-35)
𝒅𝒙 𝒅𝒙

by employing the fundamental lemma of the calculus of variations.

 This equation is normally written as follows and is called the Euler-
Poisson equation:

𝒅(𝒎) (𝒎) 𝒅(𝒎−𝟏)

𝑭 𝒚 − 𝑭𝒚 (𝒎−𝟏) + ⋯ + (−𝟏)𝒎 𝑭𝒚 = 𝟎 (8-36)
𝒅𝒙(𝒎) 𝒅𝒙(𝒎−𝟏)

This equation is an ordinary differential equation of order 2𝑚 and requires

2𝑚 boundary conditions. The following example illustrates its use in finding
the optimal function.

EXAMPLE 8-2(1)

Determine the optimum function that minimizes the integral in the following
equation:
𝒙𝟏
𝑰[𝒚(𝒙)] = ∫ [𝟏𝟔𝒚𝟐 − (𝒚′′ )𝟐 ] 𝒅𝒙
𝒙𝟎

The Euler-Poisson equation for 𝑚 = 2 is:

𝒅𝟐 𝒅
𝑭𝒚′′ − 𝑭 ′ + 𝑭𝒚 = 𝟎
𝒅𝒙 𝟐 𝒅𝒙 𝒚

Evaluating the partial derivatives gives:

𝑭 = 𝟏𝟔𝒚𝟐 − (𝒚′′ )𝟐

𝝏𝑭 𝝏𝑭 𝝏𝑭
= 𝑭𝒚′′ = −𝟐𝒚′′ = 𝑭𝒚′ = 𝟎 = 𝑭𝒚 = 𝟑𝟐𝒚
𝝏𝒚′′ 𝝏𝒚′ 𝝏𝒚

Substituting into the Euler-Poisson equation gives a fourth-order ordinary

differential equation:

𝒅𝟒 𝒚
− 𝟏𝟔𝒚 = 𝟎
𝒅𝒙𝟒

The solution of this differential equation is:

𝒚 = 𝒄𝟏 𝒆𝟐𝒙 + 𝒄𝟐 𝒆−𝟐𝒙 + 𝒄𝟑 𝒄𝒐𝒔 𝟐𝒙 + 𝒄𝟒 𝒔𝒊𝒏 𝟐𝒙

where the constants of integration are evaluated using the boundary

conditions.
Functional with Several Functions in the Integrand: For the case of the
integrand containing several functions, 𝑦1 , 𝑦2 , … , 𝑦𝑃 , the integral has the
following form:

𝒙𝟏
𝑰[𝒚𝟏 (𝒙), 𝒚𝟐 (𝒙), … , 𝒚𝑷 (𝒙)] = ∫ 𝑭[𝒙, 𝒚𝟏 , 𝒚𝟐 , … , 𝒚𝑷 , 𝒚𝟏 ′ , 𝒚𝟐 ′ , … , 𝒚𝑷 ′ ] 𝒅𝒙
𝒙𝟐
(8-37)
and boundary conditions on each of the functions are required, i.e., 𝑦1 (𝑥0 ),
𝑦1 (𝑥1 ), 𝑦2 (𝑥0 ), 𝑦2 (𝑥1 ),…, 𝑦𝑃 (𝑥0 ), 𝑦𝑃 (𝑥1 ).
The function constructed in equation (8-5) is used, except in this case 𝑝
functions are required:

̅ 𝟏 = 𝒚𝟏 + 𝜶𝟏 𝒏𝟏
𝒚
⋮ (8-38)
̅ 𝑷 = 𝒚𝑷 + 𝜶𝑷 𝒏𝑷
𝒚

with 𝑝 parameters 𝛼1 , 𝛼2 , … , 𝛼𝑃 , which can be made arbitrarily small. These

equations are substituted into equation (8-37), and then the mathematical
argument is used that the integral is a function of 𝛼1 , 𝛼2 , … , 𝛼𝑃 :

𝜱( 𝜶𝟏 , 𝜶𝟐 , … , 𝜶𝑷 )
𝒙𝟏
= ∫ 𝑭[𝒙, 𝒚𝟏 + 𝜶𝒏𝟏 , … , 𝒚𝑷 + 𝜶𝒏𝑷 , 𝒚𝟏 ′ + 𝜶𝒏𝟏 ′ , … , 𝒚𝑷 ′
𝒙𝟎
+ 𝜶𝒏𝑷 ′ ] 𝒅𝒙 (8-39)

To locate the stationary point(s) of the integral, the first partial derivatives of
Φ with respect to 𝛼1 , 𝛼2 , … , 𝛼𝑃 are set equal to zero. This gives the
following set of 𝑝 equations:

𝒙𝟏
𝝏𝜱 𝝏𝑭
=∫ 𝒅𝒙 𝒇𝒐𝒓 𝒊 = 𝟏, 𝟐, … , 𝒑 (8-40)
𝝏𝜶𝒊 𝒙𝟎 𝝏𝜶𝒊

The chain rule is used with the function 𝐹, as previously to give:

𝝏𝑭 𝝏𝑭 𝝏𝒚̅𝒊 𝝏𝑭 𝝏𝒚̅𝒊 ′ 𝝏𝑭 𝝏𝑭 ′
= + = 𝒏𝒊 + 𝒏 (8-41)
𝝏𝜶𝒊 𝝏𝒚̅𝒊 𝝏𝜶𝒊 𝝏𝒚 ′
̅𝒊 𝝏𝜶𝒊 𝝏𝒚 ̅𝒊 ̅𝒊 ′ 𝒊
𝝏𝒚

Substituting into equation (8-40), we obtain an equation comparable to

equation (8-16):
𝒙𝟏
𝝏𝜱 𝝏𝑭 𝝏𝑭 ′
=∫ [ 𝒏𝒊 + 𝒏 ] 𝒅𝒙 𝒇𝒐𝒓 𝒊 = 𝟏, 𝟐, … , 𝒑 (8-42)
𝝏𝜶𝒊 𝒙𝟎 𝝏𝒚̅𝒊 ̅𝒊 ′ 𝒊
𝝏𝒚
Integration by parts, letting 𝛼1 → 0 to have 𝑦̅𝑖 → 𝑦𝑖 and 𝑦̅𝑖 ′ → 𝑦𝑖 ′ , and to
have 𝜕𝛷⁄𝜕𝛼𝑖 = 0 gives the equation comparable to equation (8-18), i.e.:

𝒙𝟏
𝝏𝑭 𝒅 𝝏𝑭
∫ 𝒏𝒊 [ − ( )] 𝒅𝒙 𝒇𝒐𝒓 𝒊 = 𝟏, 𝟐, … , 𝒑 (8-43)
𝒙𝟎 𝝏𝒚𝒊 𝒅𝒙 𝝏𝒚𝒊 ′

Applying the fundamental lemma of the calculus of variations gives the

following set of equations comparable to equation (8-4):

𝝏𝑭 𝒅 𝝏𝑭
− ( ) = 𝟎 𝒇𝒐𝒓 𝒊 = 𝟏, 𝟐, … , 𝒑 (8-44)
𝝏𝒚𝒊 𝒅𝒙 𝝏𝒚𝒊 ′

This is a set of Euler equations, and the following example illustrates the use
of these equations to find the optimum set of functions.

EXAMPLE 8-3(1)

Determine the optimum functions that determine the stationary points for the
following integral:

𝒙𝟏
𝑰[𝒚𝟏 , 𝒚𝟐 ] = ∫ [𝟐𝒚𝟏 𝒚𝟐 − 𝟐𝒚𝟏 𝟐 + (𝒚𝟏 ′ )𝟐 − (𝒚𝟐 ′ )𝟐 ] 𝒅𝒙
𝒙𝟎

The two Euler equations are:

𝒅 𝝏𝑭 𝝏𝑭 𝒅 𝝏𝑭 𝝏𝑭
[ ]− =𝟎 [ ]− =𝟎
𝒅𝒙 𝝏𝒚𝟏 ′ 𝝏𝒚𝟏 𝒅𝒙 𝝏𝒚𝟐 ′ 𝝏𝒚𝟐

The function 𝐹 and the partial derivatives needed for these Euler equations
are:

𝑭 = 𝟐𝒚𝟏 𝒚𝟐 − 𝟐𝒚𝟏 𝟐 + (𝒚𝟏 ′ )𝟐 − (𝒚𝟐 ′ )𝟐

𝝏𝑭 𝝏𝑭
= 𝟐𝒚𝟐 − 𝟒𝒚𝟏 = 𝟐𝒚𝟏
𝝏𝒚𝟏 𝝏𝒚𝟐

𝝏𝑭 𝝏𝑭
= 𝟐𝒚𝟏 ′ = −𝟐𝒚𝟐 ′
𝝏𝒚𝟏 ′ 𝝏𝒚𝟐 ′

The two Euler equations become:

−𝒚𝟏 ′′ + 𝒚𝟐 − 𝟐𝒚𝟏 = 𝟎 𝒚𝟐 ′′ + 𝒚𝟏 = 𝟎
This set of two linear ordinary differential equations has been solved by
Forray (1), using standard techniques, and the solution is:

𝒚𝟏 = (𝒄𝟏 𝒙 + 𝒄𝟐 ) 𝐜𝐨𝐬(𝒙) + (𝒄𝟑 𝒙 + 𝒄𝟒 ) 𝐬𝐢𝐧(𝒙)

𝒚𝟐 = 𝒄𝟏 (𝒙 𝐜𝐨𝐬(𝒙) − 𝟐 𝐬𝐢𝐧(𝒙)) + 𝒄𝟐 𝐜𝐨𝐬(𝒙) + 𝒄𝟑 (𝟐 𝐜𝐨𝐬(𝒙) + 𝒙 𝐬𝐢𝐧(𝒙))

+ 𝒄𝟒 𝐬𝐢𝐧(𝒙)

Functional with Several Functions and Higher Derivatives: We can now

consider the case that combines the two previous ones, i.e., the integrand
contains several functions with higher order derivatives:
𝒙𝟏
𝑰[𝒚𝟏 , … , 𝒚𝑷 ] = ∫ 𝑭[𝒙, 𝒚𝟏 , 𝒚𝟏 ′ , … , 𝒚𝟏 (𝒎) , … , 𝒚𝑷 , 𝒚𝒑 ′ , … , 𝒚𝑷 (𝒌) ] 𝒅𝒙 (8-45)
𝒙𝟎

The procedure to obtain the set of ordinary differential equations to

determine the stationary points is a combination of the derivations for the
two previous cases. The integral is converted to a function of parameters 𝛼1 ,
𝛼2 , … , 𝛼𝑃 , and the first partial derivatives with respect to these parameters
are set equal to zero to give the following set of Euler-Poisson equations:

𝒅 𝒅
(𝒎)
𝑭𝒚𝟏 (𝒎) − (𝒎−𝟏) 𝑭𝒚𝟏 (𝒎−𝟏) + ⋯ + (−𝟏)𝒎 𝑭𝒚𝟏 = 𝟎
𝒅𝒙 𝒅𝒙
⋮ (8-46)
(𝒌) (𝒌−𝟏)
𝒅 𝒅
𝑭𝒚 (𝒌) − (𝒌−𝟏) 𝑭𝒚𝑷 (𝒌−𝟏) + ⋯ + (−𝟏)𝒌 𝑭𝒚𝑷 = 𝟎
𝒅𝒙(𝒌) 𝑷 𝒅𝒙

This is a set of 𝑝 ordinary differential equations, and the order of each one
is determined by the highest order derivative appearing in the integrand. The
following example illustrates the use of these equations.

EXAMPLE 8-4

Determine the optimal functions that determine the stationary points for the
following integral:
𝒙𝟏
𝑰[𝒚𝟏 , 𝒚𝟐 ] = ∫ {[𝟏 + (𝒚𝟏 ′ )𝟐 ]𝟏⁄𝟐 + 𝟏𝟔𝒚𝟐 𝟐 − (𝒚𝟐 ′′ )𝟐 } 𝒅𝒙
𝒙𝟎

The Euler-Poisson equations for 𝑦1 (𝑚 = 1) and 𝑦2 (𝑘 = 2)

𝒅 𝒅𝟐 𝒅𝑭𝒚𝟐 ′
𝑭 ′ − 𝑭𝒚𝟏 = 𝟎 𝑭 ′′
− + 𝑭𝒚𝟐 = 𝟎
𝒅𝒙 𝒚𝟏 𝒚
𝒅𝒙𝟐 𝟐 𝒅𝒙
Computing the partial derivatives gives:

𝑭𝒚𝟏′ = 𝒚𝟏 ⁄[𝟏 + (𝒚𝟏 )𝟐 ]𝟑⁄𝟐 𝑭𝒚𝟐 ′′ = −𝟐𝒚𝟐 ′′

𝑭 𝒚𝟏 = 𝟎 𝑭𝒚𝟐′ = 𝟎

𝑭𝒚𝟐 = 𝟑𝟐𝒚𝟐

substituting gives

𝒅 𝒚𝟏 ′ 𝒅
[ ]=𝟎 (−𝟐𝒚𝟐 ′′ ) + 𝟑𝟐𝒚𝟐 = 𝟎
𝒅𝒙 [𝟏 + (𝒚𝟏 ′ )𝟐 ]𝟑⁄𝟐 𝒅𝒙𝟐

These are two ordinary differential equations in 𝑦1 and 𝑦2 . The solution to

each differential equation is given in Examples 8-1 and 8-2. In fact, the
example was constructed from these two problems for a simple illustration
of the application of a set of Euler-Poisson equations. However, it does not
illustrate the coupling that would normally occur, which requires the set of
equations to be solved simultaneously.

An example of this coupling is given in the outline of the optimal rocket

trajectory problem by Wylie and Barrett (8), which requires a solution of two
Euler-Poisson equations. The equation for the conservation of momentum is
applied to the rocket, and the initial conditions on the position of the rocket
and the rate of fuel use are required information to determine a family of
optimal trajectories.

Functional with More than One Independent Variable: For this case the
integrand contains more than one independent variable. The analogous form
to equation (8-2) for two independent variables is:

𝑰[𝒚(𝒙𝟏 , 𝒙𝟐 )] = ∫ ∫ 𝑭(𝒙𝟏 , 𝒙𝟐 , 𝒚, 𝒚𝒙𝟏 , 𝒚𝒙𝟐 ) 𝒅𝒙𝟏 𝒅𝒙𝟐 (8-47)

𝑹

where the integral is integrated over the region 𝑅, and 𝑦𝑥1 and 𝑦𝑥2 indicate
partial differentiation of 𝑦 with respect to 𝑥1 and 𝑥2 .
The procedure to obtain the differential equation to be solved for the
optimal solution of equation (8-47) follows the mathematical arguments used
for the case of one independent variable. However, Green's theorem is
required for the integration by parts, and the function 𝑛(𝑥1 , 𝑥2 ) is zero on the
surface
of the region 𝑅. The function 𝑦̅(𝑥1 , 𝑥2 ) is constructed from the optimal
function 𝑦(𝑥1 , 𝑥2 ) and the arbitrary function 𝑛(𝑥1 , 𝑥2 ) as:

̅(𝒙𝟏 , 𝒙𝟐 ) = 𝒚(𝒙𝟏 , 𝒙𝟐 ) + 𝜶𝒏(𝒙𝟏 , 𝒙𝟐 )

𝒚 (8-48)

where 𝛼 is the parameter that can be made arbitrarily small.

Now the integral in equation (8-47) can be considered to be a function of
𝛼 only, as was done in the previous mathematical arguments, i.e.;

̅(𝒙𝟏 , 𝒙𝟐 )] = 𝑰[𝒚(𝒙𝟏 , 𝒙𝟐 ) + 𝜶𝒏(𝒙𝟏 , 𝒙𝟐 )] = 𝜱(𝜶)

𝑰[𝒚 (8-49)

Then differentiating with respect to 𝛼 gives an equation comparable to

equation (8-12). The surface of the region 𝑅 is a constant allowing the
interchange of the order of differentiation and integration:

𝒅𝜱 𝒅
=∫ ∫ 𝑭(𝒙𝟏 , 𝒙𝟐 , 𝒚, 𝒚𝒙𝟏 , 𝒚𝒙𝟐 ) 𝒅𝒙𝟏 𝒅𝒙𝟐 (8-50)
𝒅𝜶 𝑹 𝒅𝜶

Applying the chain rule, as was done previously where 𝐹 is not considered a
function of 𝑥1 and 𝑥2 for changes from surface to surface, the integrand
becomes:

̅
𝒅𝑭 𝝏𝑭 𝝏𝒚 ̅𝒙𝟏
𝝏𝑭 𝝏𝒚 ̅𝒙𝟐
𝝏𝑭 𝝏𝒚
= + + (8-51)
̅ 𝝏𝜶 𝝏𝒚
𝒅𝜶 𝝏𝒚 ̅𝒙𝟏 𝝏𝜶 ̅𝒙𝟐 𝝏𝜶
𝝏𝒚

and

̅
𝝏𝒚 ̅𝒙𝟏
𝝏𝒚 𝝏𝒏 ̅𝒙𝟐
𝝏𝒚 𝝏𝒏
=𝒏 = =
𝝏𝜶 𝝏𝜶 𝝏𝒙𝟏 𝝏𝜶 𝝏𝒙𝟐

The integral in equation (8-50) can be written in the following form, using
equation (8-51):

𝒅𝜱 𝝏𝑭 𝝏𝑭 𝝏𝒏 𝝏𝑭 𝝏𝒏
= ∫ ∫[ 𝒏 + + ] 𝒅𝒙𝟏 𝒅𝒙𝟐 (8-52)
𝒅𝜶 𝑹 ̅
𝝏𝒚 ̅
𝝏𝒚 𝒙𝟏 𝝏𝒙 𝟏 ̅𝒙𝟐 𝝏𝒙𝟐
𝝏𝒚

which is comparable to equation (8-16).

In this case the integration-by-parts is performed using Green's theorem
in the plane, which is:

𝝏𝒇 𝝏𝒇
∫ ∫ [𝑮 +𝑯 ] 𝒅𝒙𝟏 𝒅𝒙𝟐 (8-53)
𝑫 𝝏𝒙𝟏 𝝏𝒙𝟐
 𝝏𝑮 𝝏𝑯
= −∫ ∫𝒇[ + ] 𝒅𝒙𝟏 𝒅𝒙𝟐 + ∫ 𝒇(𝑮𝒅𝒙𝟏 − 𝑯𝒅𝒙𝟐 )
𝑫 𝝏𝒙𝟏 𝝏𝒙𝟐 𝑪

This theorem is applied to the second two terms of equation (8-52), where
𝑓 = 𝑛, 𝐺 = 𝜕𝐹 ⁄𝜕𝑦𝑥1 and 𝐻 = 𝜕𝐹 ⁄𝜕𝑦𝑥2 . An equation comparable to
equation (8-18) is obtained by allowing 𝛼 → 0, such that 𝑦̅ → 𝑦, 𝑦̅𝑥1 → 𝑦𝑥1
and 𝑦̅𝑥2 → 𝑦𝑥2 ; and 𝑑𝛷⁄𝑑𝛼 = 0 to have:

𝝏𝑭 𝝏 𝝏𝑭 𝝏 𝝏𝑭
∫ ∫𝒏[ − ( )− ( )] 𝒅𝒙𝟏 𝒅𝒙𝟐 = 𝟎 (8-54)
𝑹 𝝏𝒚 𝝏𝒙𝟏 𝝏𝒚𝒙𝟏 𝝏𝒙𝟐 𝝏𝒚𝒙𝟐

Again, it is argued, using an extension of the fundamental lemma of the

calculus of variations, that if the integral is equal to zero, then the term in the
brackets is equal to zero, because 𝑛(𝑥1 , 𝑥2 ) is arbitrary everywhere except
on the boundaries where it is zero. The result is the equation that corresponds
to the Euler equation, equation (8-4):

𝝏𝑭 𝝏 𝝏𝑭 𝝏 𝝏𝑭
− ( )− ( )=𝟎 (8-55)
𝝏𝒚 𝝏𝒙𝟏 𝝏𝒚𝒙𝟏 𝝏𝒙𝟐 𝝏𝒚𝒙𝟐

Also, this equation can be expanded using the chain rule to give an equation
that corresponds to equation (8-21), which is:

𝝏𝟐 𝒚 𝝏𝟐 𝒚 𝝏𝟐 𝒚 𝝏𝒚 𝝏𝒚
𝑭𝒚𝒙 𝒚 + 𝟐𝑭 𝒚 𝒙𝟏 𝒚 𝒙𝟐 + 𝑭𝒚 𝒙𝟐 𝒚 𝒙𝟐 + 𝑭𝒚𝒙 𝒚 + 𝑭𝒚𝒙 𝒚
𝟏 𝒙𝟏 𝝏𝒙 𝟐 𝝏𝒙𝟏 𝝏𝒙𝟐 𝝏𝒙𝟐 𝟐 𝟏 𝝏𝒙𝟏 𝟐 𝝏𝒙𝟐
𝟏
+ 𝑭𝒚𝒙 𝒙𝟏 + 𝑭𝒚𝒙 𝒙𝟐 − 𝑭𝒚 = 𝟎 (8-56)
𝟏 𝟐

It can be seen that this is a second-order partial differential equation in

two independent variables. Appropriate boundary conditions at the surface
in terms of 𝑦 and the first partial derivatives of 𝑦 are required for a solution.
Prior to illustrating the use of equation (8-55), a general form is given
from Burley (9) by the following equation for 𝑛 independent variables with
only first partial derivatives in the integrand:
𝒏
𝝏 𝝏𝑭 𝝏𝑭
∑ ( )− =𝟎 (8-57)
𝝏𝒙𝒊 𝝏𝒚𝒙𝒊 𝝏𝒚
𝒊=𝟏

The derivation of this equation follows the one for two independent
variables.
 The following example illustrates an application of equation (8-55). Other
applications are given by Forray (1) and Schechter (4).

EXAMPLE 8-5 (1)

The following equation describes the potential energy of a stretched

membrane, which is a minimum for small deflections. If 𝐴 is the tension per
unit length and 𝐵 is the external load, the optimum shape 𝑦(𝑥1 , 𝑥2 ) is
determined by minimizing the integral:

𝝏𝒚 𝟐 𝝏𝒚 𝟐
𝑰 = 𝟏⁄𝟐 ∫ ∫ {𝑨 ( ) + 𝑨( ) − 𝟐𝑩𝒚} 𝒅𝒙𝟏 𝒅𝒙𝟐
𝑫 𝝏𝒙𝟏 𝝏𝒙𝟐

Obtain the differential equation that is to be solved for the optimum shape.
The extension of the Euler equation for this case of two independent
variables was given by equation (8-55):

𝝏 𝝏𝑭 𝝏 𝝏𝑭 𝝏𝑭
( )+ ( )− =𝟎
𝝏𝒙𝟏 𝝏𝒚𝒙𝟏 𝝏𝒙𝟐 𝝏𝒚𝒙𝟐 𝝏𝒚

The integrand for 𝐹 in the above equation is:

𝝏𝒚 𝟐 𝝏𝒚 𝟐
𝑭 = 𝑨( ) + 𝑨( ) − 𝟐𝑩𝒚
𝝏𝒙𝟏 𝝏𝒙𝟐

The following results are obtained from evaluating the partial derivatives:

𝝏𝑭 𝝏𝒚 𝝏𝑭 𝝏𝒚 𝝏𝑭
= 𝟐𝑨 = 𝟐𝑨 = −𝟐𝑩
𝝏𝒚𝒙𝟏 𝝏𝒙𝟏 𝝏𝒚𝒙𝟐 𝝏𝒙𝟐 𝝏𝒚

Substituting into the equation and simplifying gives:

𝝏𝟐 𝒚 𝝏𝟐 𝒚 𝑩
𝟐
+ + =𝟎
𝝏𝒙𝟏 𝝏𝒙𝟐 𝟐 𝑨

This is a second-order, elliptic partial differential equation. The solution

requires boundary conditions that give the shape of the sides of the
membrane. The solution of the partial differential equation will be the shape
of the membrane.
The text by Courant and Hilbert (5) gives the extension for higher order
derivatives in the integrand. Also, that book gives solutions for a number
of problems, including membrane shapes of rectangles and circles, and is an
excellent reference book on free and forced vibrations of membranes.
In the next section, the results for unconstrained problems are extended
to those with constraints. The constraints can be of three types for calculus
of variations problems, and they are algebraic, integral, and differential
equations.

CONSTRAINED VARIATIONAL PROBLEMS

Generally, there are two procedures used for solving variational problems
that have constraints. These are the methods of direct substitution and
Lagrange Multipliers. In the method of direct substitution, the constraint
equation is substituted into the integrand; and the problem is converted into
an unconstrained problem, as was done in Chapter 2. In the method of
Lagrange Multipliers, the Lagrangian function is formed, and the
unconstrained problem is solved using the appropriate forms of the Euler or
Euler-Poisson equation. However, in some cases the Lagrange Multiplier is
a function of the independent variables and is not a constant. This is an added
complication that was not encountered in Chapter 2.

Algebraic Constraints: To illustrate the method of Lagrange Multipliers,

the simplest case with one algebraic equation will be used. The extension to
more complicated cases is the same as that for analytical methods:
𝒙𝟏
𝑶𝒑𝒕𝒊𝒎𝒊𝒛𝒆: 𝑰[𝒚(𝒙)] = ∫ 𝑭(𝒙, 𝒚, 𝒚′ ) 𝒅𝒙 (8-58)
𝒙𝟎

𝑺𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐: 𝑮(𝒙, 𝒚) = 𝟎

The Lagrangian function is formed, as shown below:

𝑳(𝒙, 𝒚, 𝒚′ , 𝝀) = 𝑭(𝒙, 𝒚, 𝒚′ ) + 𝝀(𝒙)𝑮(𝒙, 𝒚) (8-59)

The Lagrange Multiplier 𝜆 is a function of the independent variable, 𝑥,

and the unconstrained Euler equation is solved as given below:

𝒅 𝝏𝑳 𝝏𝑳
( )− =𝟎 (8-60)
𝒅𝒙 𝝏𝒚′ 𝝏𝒚

along with the constraint equation 𝐺(𝑥, 𝑦) = 0.

There is a Lagrange Multiplier for each constraint equation when the
Lagrangian function is formed. A derivation of the Lagrange Multiplier
method is given by Forray (1), and the following example illustrates the
technique.

EXAMPLE 8-6 (8)

The classic example to illustrate this procedure is the problem of finding the
path of a unit mass particle on a sphere from point (0,0,1) to point (0,0, −1)
in time 𝑇, which minimizes the integral of the kinetic energy of the particle.
The integral to be minimized and the constraint to be satisfied are:

𝑻
𝑴𝒊𝒏𝒊𝒎𝒊𝒛𝒆: 𝑰[𝒙, 𝒚, 𝒛] = ∫ [(𝒙′ )𝟐 + (𝒚′ )𝟐 + (𝒛′ )𝟐 ]𝟏⁄𝟐 𝒅𝒕
𝟎

𝑺𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐: 𝒙𝟐 + 𝒚𝟐 + 𝒛𝟐 = 𝟏

The Lagrangian function is:

𝑳[𝒙(𝒕), 𝒚(𝒕), 𝒛(𝒕), 𝝀(𝒕)] = [(𝒙′ )𝟐 + (𝒚′ )𝟐 + (𝒛′ )𝟐 ]𝟏⁄𝟐 + 𝝀(𝒙𝟐 + 𝒚𝟐 + 𝒛𝟐 − 𝟏)

There are three optimal functions to be determined, and the corresponding

three Euler equations are:

𝒅 𝝏𝑳 𝝏𝑳 𝒅 𝝏𝑳 𝝏𝑳 𝒅 𝝏𝑳 𝝏𝑳
( ′) − =𝟎 ( ′) − =𝟎 ( ′) − =𝟎
𝒅𝒕 𝝏𝒙 𝝏𝒙 𝒅𝒕 𝝏𝒚 𝝏𝒚 𝒅𝒕 𝝏𝒛 𝝏𝒛

Performing the partial differentiation of 𝐿, recognizing that [(𝑥 ′ )2 + (𝑦 ′ )2 +

(𝑧 ′ )2 ]1⁄2 = 𝑠 ′ (an arc length that is a constant), and substituting into the
Euler equations, we obtain three simple, second-order ordinary differential
equations:

𝒅𝟐 𝒙 𝒅𝟐 𝒚 𝒅𝟐 𝒛
+ 𝝀𝒙 = 𝟎 + 𝝀𝒚 = 𝟎 + 𝝀𝒛 = 𝟎
𝒅𝒕𝟐 𝒅𝒕𝟐 𝒅𝒕𝟐

These can be integrated after some manipulations to give:

𝒙 + 𝒄𝟏 𝒚 + 𝒄𝟐 𝒛 = 𝟎

which is the equation of a plane through the center of the sphere. The
intersection of this plane and the sphere is a great circle, which is the optimal
path. It can be shown that the minimum kinetic energy is 𝜋 2 ⁄𝑇.
Integral Constraints: Isoperimetric problems (1) are ones where an integral
is to be optimized subject to a constraint, which is another integral having a
specified value. This name came from the famous problem of Dido of finding
the closed curve of given perimeter for which the area is a maximum. For the
Euler equation the problem can be stated as:

𝒙𝟏
𝑶𝒑𝒕𝒊𝒎𝒊𝒛𝒆: 𝑰[𝒚(𝒙)] = ∫ 𝑭(𝒙, 𝒚, 𝒚′ ) 𝒅𝒙
𝒙𝟎

𝒙𝟏
(8-61)
′)
𝑺𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐: 𝑱 = ∫ 𝑮(𝒙, 𝒚, 𝒚 𝒅𝒙
𝒙𝟎

where 𝐽 is a known constant.

To solve this problem, the Lagrangian function 𝐿(𝑥, 𝑦, 𝑦 ′ ) is formed as
shown below:

𝑳(𝒙, 𝒚, 𝒚′ , 𝝀) = 𝑭(𝒙, 𝒚, 𝒚′ ) + 𝝀𝑮(𝒙, 𝒚, 𝒚′ ) (8-62)

and the following unconstrained Euler equation is solved along with the
constraint equation:

𝒅 𝝏𝑳 𝝏𝑳
( ′) − =𝟎 (8-60)
𝒅𝒙 𝝏𝒚 𝝏𝒚

For integral equation constraints, the Lagrange multiplier 𝜆 is a constant,

and each constraint has a Lagrange multiplier when forming the Lagrangian
function. The following example illustrates the use of Lagrange multipliers
with an integral constraint. It is the classic problem of Dido mentioned
previously.

EXAMPLE 8-7 (1)

Determine the shape of the curve of length 𝐽 that encloses the maximum area.
The integral to be maximized and the integral constraint are as follows:

𝒙𝟏
𝑴𝒂𝒙𝒊𝒎𝒊𝒛𝒆: 𝑰[𝒚(𝒙)] = 𝟏⁄𝟐 ∫ [𝒚𝟏 𝒚𝟐 ′ − 𝒚𝟐 𝒚𝟏 ′ ]𝒅𝒙
𝒙𝟎

𝒙𝟏
𝟐 𝟐 𝟏⁄𝟐
𝑺𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐: 𝑱 = ∫ [𝒚𝟏 ′ + 𝒚𝟐 ′ ] 𝒅𝒙
𝒙𝟎
The Lagrangian function is:

𝟐 𝟐 𝟏⁄𝟐
𝑳 = 𝒚𝟏 𝒚𝟐 ′ − 𝒚𝟐 𝒚𝟏 ′ + 𝝀[𝒚𝟏 ′ + 𝒚𝟐 ′ ]

The two Euler equations are:

𝒅 𝝏𝑳 𝝏𝑳 𝒅 𝝏𝑳 𝝏𝑳
( ′
)− =𝟎 ( ′
)− =𝟎
𝒅𝒙 𝝏𝒚𝟏 𝝏𝒚𝟏 𝒅𝒙 𝝏𝒚𝟐 𝝏𝒚𝟐

Performing the differentiation and substituting into the Euler equations give:

𝒅 𝟐 𝟐 −𝟏⁄𝟐
{−𝒚𝟐 + 𝒚𝟏 ′ 𝝀[𝒚𝟏 ′ + 𝒚𝟐 ′ ] } − 𝒚𝟐 ′ = 𝟎
𝒅𝒙

𝒅 𝟐 𝟐 −𝟏⁄𝟐
{𝒚 + 𝒚𝟐 ′ 𝝀[𝒚𝟏 ′ + 𝒚𝟐 ′ ] } − 𝒚𝟏 ′ = 𝟎
𝒅𝒙 𝟏

The two equations above can be integrated once to obtain the following
results:

−𝒚𝟏 ′ 𝝀 −𝒚𝟐 ′ 𝝀
𝒚 𝟐 − 𝒄𝟐 = 𝟏⁄𝟐
𝒚 𝟏 − 𝒄𝟏 = 𝟏⁄𝟐
𝟐[𝒚𝟏 ′ 𝟐 + 𝒚𝟐 ′ 𝟐 ] 𝟐[𝒚𝟏 ′ 𝟐 + 𝒚𝟐 ′ 𝟐 ]

Squaring both sides and adding the two equations gives the following:

(𝒚𝟏 − 𝒄𝟏 )𝟐 + (𝒚𝟐 − 𝒄𝟐 )𝟐 = 𝝀𝟐 ⁄𝟒 = 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕

which is the equation of a circle. Thus, a circle encloses the maximum area
for a given length curve.

Differential Equation Constraints: To illustrate the method of Lagrange

Multipliers for differential equation constraints, a simple case will be used.
Extensions to more detailed cases are the same as for the two previous types
of constraints. The problem is as follows:
𝒙𝟏
𝑶𝒑𝒕𝒊𝒎𝒊𝒛𝒆: 𝑰[𝒚(𝒙)] = ∫ 𝑭(𝒙, 𝒚, 𝒚′ ) 𝒅𝒙
𝒙𝟎
(8-63)
𝑺𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐: 𝑮(𝒙, 𝒚, 𝒚′ ) = 𝟎

As was done previously, the Lagrangian function is formed as follows:

 𝑳(𝒙, 𝒚, 𝒚′ , 𝝀) = 𝑭(𝒙, 𝒚, 𝒚′ ) + 𝝀(𝒙)𝑮(𝒙, 𝒚, 𝒚′ ) (8-64)

Then the Lagrangian function is used in the Euler equation:

𝒅 𝝏𝑳 𝝏𝑳
( ′) − =𝟎 (8-60)
𝒅𝒙 𝝏𝒚 𝝏𝒚

In this case the Lagrange Multiplier 𝜆(𝑥) is a function of the independent

variable. This procedure is illustrated in the following example, which was
given by Beveridge and Schechter (10). Also, they extend these results to
obtain Pontryagin's maximum principle for constraints placed on the range
of the dependent and independent variables.

EXAMPLE 8-8 (10)

The following problem to minimize 𝐼[𝑦1 , 𝑦2 ] has a differential equation

constraint:
𝒙𝟏
𝑰[𝒚𝟏 (𝒙), 𝒚𝟐 (𝒙)] = ∫ (𝒚𝟏 𝟐 + 𝒚𝟐 𝟐 ) 𝒅𝒙
𝒙𝟎

subject to:

𝒅𝒚𝟏
= 𝒚 𝟐 − 𝒚𝟏
𝒅𝒙

The Lagrangian function is:

𝑳 = 𝒚𝟏 𝟐 + 𝒚𝟐 𝟐 + 𝝀(𝒚𝟏 ′ − 𝒚𝟐 + 𝒚𝟏 )

Using equation (8-60) obtains the two Euler equations for 𝑦1 and 𝑦2 . They
are to be solved with the constraint equation, and this gives the following set
of equations:

−𝟐𝒚𝟏 − 𝝀 + 𝝀′ = 𝟎

𝟐𝒚𝟐 − 𝝀 = 𝟎

𝒚𝟏 ′ + 𝒚𝟏 − 𝒚𝟐 = 𝟎

The solutions for 𝑦1 and 𝑦2 are obtained by manipulating and integrating the
equation set to give:
 𝒚𝟏 = 𝒄𝟏 𝒆√𝟐𝒙 + 𝒄𝟐 𝒆√𝟐𝒙

𝒚𝟐 = 𝒄𝟏 (𝟏 + √𝟐)𝒆√𝟐𝒙 + 𝒄𝟐 (𝟏 − √𝟐)𝒆−√𝟐𝒙

where the constants of integration 𝑐1 and 𝑐2 are evaluated using the boundary
conditions. A particular solution for 𝑦1 (0) = 1 and 𝑦2 (𝑥1 ) = 0 is given by
Beveridge and Schechter (10).

The previous examples were designed to illustrate the particular

extension of the calculus of variations and were essentially simple
mathematics problems with no industrial application associated with them.
However, the following example was designed to illustrate the application
of the calculus of variations to a process, and it employs unsteady material
and energy balance equations to determine the optimum way to control the
flow rate to an agitated tank. Although the example is relatively simple, it
illustrates economic model and process constraints for a dynamic system;
and an optimal control function is developed.

EXAMPLE 8-9 (11)

An agitated tank contains 𝑊 pounds of water at 32 °𝐹. It is desired to raise

the temperature of the water in the tank to 104 °𝐹 in (2.0)1⁄2 hours by
feeding water at a rate of 𝑊 pounds per hour. The tank is completely filled
with water, and the overflow water at 𝑇2 (𝑡) is equal to the input flow rate at
𝑇1 (𝑡). The average residence time of water in the tank is 1 ℎ𝑜𝑢𝑟, and the
tank is perfectly mixed. The temperature of the inlet can be adjusted as a
function of time by an electric heater in the feed pipe, which is connected to
a variable voltage transformer. The sensible heat accompanying water
flowing into and out of the tank during the process must be considered lost.
Therefore, it is, desired to minimize the integral of the sum of squares of the
difference between the temperatures, 𝑇1 (𝑡) and 𝑇2 (𝑡), and the reference
temperature, 32 °𝐹. This economic model is given by the following equation:

√𝟐
𝑰[𝑻𝟏 (𝒕), 𝑻𝟐 (𝒕)] = ∫ {[𝑻𝟏 (𝒕) − 𝟑𝟐]𝟐 + [𝑻𝟐 (𝒕) − 𝟑𝟐]𝟐 } 𝒅𝒕
𝟎

An unsteady-state energy balance on the water in the tank at time 𝑡 gives the
following equation relating the temperatures 𝑇1 (𝑡), 𝑇2 (𝑡) and the system
parameters:
𝒅
𝑪𝑷 𝑾 𝑻 (𝒕) = 𝑾𝑪𝑷 [𝑻𝟏 (𝒕) − 𝟑𝟐] − 𝑾𝑪𝑷 [𝑻𝟐 (𝒕) − 𝟑𝟐]
𝒅𝒕 𝟐
For water the heat capacity, 𝐶𝑃 , is equal to 1.0 𝐵𝑇𝑈⁄𝑙𝑏℉, and this equation
simplifies to the following form:

𝒅
𝑻 (𝒕) = 𝑻𝟏 (𝒕) − 𝑻𝟐 (𝒕)
𝒅𝒕 𝟐

The calculus of variations problem can now be formulated as:

√𝟐
𝑴𝒊𝒏𝒊𝒎𝒊𝒛𝒆: 𝑰[𝑻𝟏 (𝒕), 𝑻𝟐 (𝒕)] = ∫ {[𝑻𝟏 (𝒕) − 𝟑𝟐]𝟐 + [𝑻𝟐 (𝒕) − 𝟑𝟐]𝟐 } 𝒅𝒕
𝟎

𝒅
𝑺𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐: 𝑻 (𝒕) − 𝑻𝟏 (𝒕) + 𝑻𝟐 (𝒕) = 𝟎
𝒅𝒕 𝟐

with 𝑇1 (0) = 𝑇2 (0) = 32 ℉, and 𝑇2 (√2) = 104 ℉, as boundary

conditions.
Two optimal functions are determined, and the solution of two Euler
equations is required, using equation (8-60). The Lagrangian function is:

𝑳[𝑻𝟏 (𝒕), 𝑻𝟐 (𝒕), 𝝀(𝒕)] = (𝑻𝟏 − 𝟑𝟐)𝟐 + (𝑻𝟐 − 𝟑𝟐)𝟐 + 𝝀(𝒕)[𝑻𝟐 ′ − 𝑻𝟏 + 𝑻𝟐 ]

and the Euler equations are:

𝒅 𝝏𝑳 𝝏𝑳 𝒅 𝝏𝑳 𝝏𝑳
( )− =𝟎 ( )− =𝟎
𝒅𝒕 𝝏𝑻𝟏 ′ 𝝏𝑻𝟏 𝒅𝒕 𝝏𝑻𝟐 ′ 𝝏𝑻𝟐

The results of performing the differentiation are:

𝝏𝑳 𝝏𝑳
= 𝟐(𝑻𝟏 − 𝟑𝟐) − 𝝀 = 𝟐(𝑻𝟐 − 𝟑𝟐) + 𝝀
𝝏𝑻𝟏 𝝏𝑻𝟐

𝝏𝑳 𝝏𝑳 𝒅 𝝏𝑻 𝒅 𝝏𝑻 𝒅𝝀
=𝟎 = 𝝀(𝒕) ( )=𝟎 ( ′) =
𝝏𝑻𝟏 ′ 𝝏𝑻𝟐 ′ 𝒅𝒕 𝝏𝑻𝟏 ′ 𝒅𝒕 𝝏𝑻𝟐 𝒅𝒕

𝒅 𝝏𝑳 𝒅 𝝏𝑳 𝒅𝝀
( )=𝟎 ( )=
𝒅𝒕 𝝏𝑻𝟏 ′ 𝒅𝒕 𝝏𝑻𝟐 ′ 𝒅𝒕

Substituting into the Euler equations gives the following set of equations:

𝟐(𝑻𝟏 − 𝟑𝟐) − 𝝀 = 𝟎
 𝒅𝝀
𝟐(𝑻𝟐 − 𝟑𝟐) + 𝝀 − =𝟎
𝒅𝒕

𝒅
𝑻 − 𝑻𝟏 + 𝑻𝟐 = 𝟎
𝒅𝒕 𝟐

The third equation is the constraint, and these equations are solved for 𝑇1 (𝑡)
and 𝑇2 (𝑡). The set has two ordinary differential equations and one algebraic
equation. Manipulating and solving this set for one equation in terms of 𝑇2 (𝑡)
gives:

𝑻𝟐 ′′ − 𝟐𝑻𝟐 = −𝟔𝟒

With the boundary conditions of 𝑇2 (0) = 32 and 𝑇2 (√2) = 104, the

solution to the differential equation is:

𝑻𝟐 (𝒕) = 𝟗. 𝟗𝟏[𝒆√𝟐𝒕 − 𝒆−√𝟐𝒕 ] + 𝟑𝟐

where 9.91 = 72⁄(𝑒 2 − 𝑒 −2 ). The constraint is used to obtain the entering

water temperature as a function of time, and substituting in the solution for
𝑇2 (𝑡) gives:

𝑻𝟏 (𝒕) = 𝟗. 𝟗𝟏[(𝟐. 𝟒𝟏𝟒𝟒)𝒆√𝟐𝒕 + (𝟎. 𝟒𝟏𝟒𝟒)𝒆−√𝟐𝒕 ] + 𝟑𝟐

The solutions for the optimal functions, 𝑇1 (𝑡) and 𝑇2 (𝑡) are tabulated and
plotted in Figure 8-3. As shown in the figure, the warm water temperature
increases to 209 °𝐹 for the water temperature in the tank to reach 104 °𝐹 in
√2 ℎ𝑜𝑢𝑟𝑠.

CLOSURE

Some of the important results from the chapter are summarized in an

abbreviated form in Table 8-1. First, a set of Euler equations is shown in the
table to be solved when the integrand contains several optimal functions and
their first derivatives. Corresponding boundary conditions are required on
each of these Euler equations, which are second-order ordinary differential
equations. Next in the table is the integral that has higher order derivatives
in the integrand. For this case the Euler-Poisson equation has to be solved,
and it is of order 2𝑚, where 𝑚 is the order of the highest derivative in the
integrand.
Also, appropriate boundary conditions on 𝑦 and its derivatives at 𝑥0 and
Figure 8-3. Optimal temperature functions for agitated tank application of
Example 8-9 (11).

𝑥1 are required to obtain the particular solution of the differential equation.

A combination of these two cases is given in the table where a set of Euler-
Poisson equations is solved for the optimum functions.
When the optimal function involves more than one independent variable,
a partial differential equation has to be solved, and the table shows the case
for two independent variables, a second-order partial differential equation.
Equation (8-57) gives the comparable equation for 𝑛 independent variables.
However, the results given in the table and the chapter are only for an optimal
function with first partial derivatives in the integrand. Results comparable
 Table 8-1. Summary of Results for the Calculus of Variations.
1. Optimize:
𝒙𝟏
𝑰[𝒚𝟏 (𝒙), 𝒚𝟐 (𝒙), … , 𝒚𝑷 (𝒙)] = ∫ 𝑭[𝒙, 𝒚𝟏 , 𝒚𝟐 , … , 𝒚𝑷 , 𝒚𝟏 ′ , 𝒚𝟐 ′ , … , 𝒚𝑷 ′ ] 𝒅𝒙
𝒙𝟐
Solve a set of 𝑝, second-order ordinary differential equations:
𝝏𝑭 𝒅 𝝏𝑭
− ( ) = 𝟎 𝒇𝒐𝒓 𝒊 = 𝟏, 𝟐, … , 𝒑
𝝏𝒚𝒊 𝒅𝒙 𝝏𝒚𝒊 ′

2. Optimize:
𝒙𝟏
𝑰[𝒚(𝒙)] = ∫ 𝑭[𝒙, 𝒚, 𝒚′ , … , 𝒚(𝒎) ] 𝒅𝒙
𝒙𝟎
Solve an ordinary differential equation of order 2𝑚:

𝒅(𝒎) 𝝏𝑭 𝒅(𝒎−𝟏) 𝝏𝑭 𝝏𝑭
( ) − (𝒎−𝟏) ( (𝒎−𝟏) ) + ⋯ + (−𝟏)𝒎 =𝟎
𝒅𝒙(𝒎) 𝝏𝒚(𝒎) 𝒅𝒙 𝝏𝒚 𝝏𝒚

3. Optimize:
𝒙𝟏
𝑰[𝒚𝟏 , 𝒚𝟐 , … , 𝒚𝑷 ] = ∫ 𝑭[𝒙, 𝒚𝟏 , 𝒚𝟏 ′ , … , 𝒚𝟏 (𝒎) , … , 𝒚𝑷 , 𝒚𝒑 ′ , … , 𝒚𝑷 (𝒌) ] 𝒅𝒙
𝒙𝟎
Solve a set of 𝑝 Euler-Poisson equations:
𝒅(𝒎) 𝝏𝑭 𝒅(𝒎−𝟏) 𝝏𝑭 𝝏𝑭
(𝒎) ( (𝒎) ) − ( ) + ⋯ + (−𝟏)𝒎 =𝟎
𝒅𝒙 𝝏𝒚𝟏 𝒅𝒙(𝒎−𝟏) 𝝏𝒚𝟏 (𝒎−𝟏) 𝝏𝒚𝟏
⋮
𝒅(𝒌) 𝝏𝑭 𝒅(𝒌−𝟏) 𝝏𝑭 𝝏𝑭
( ) − (𝒌−𝟏) ( (𝒌−𝟏) )
+ ⋯ + (−𝟏)𝒌 =𝟎
𝒅𝒙(𝒌) 𝝏𝒚𝑷 (𝒌) 𝒅𝒙 𝝏𝒚𝑷 𝝏𝒚𝑷

4. Optimize:
𝑰[𝒚(𝒙𝟏 , 𝒙𝟐 )] = ∫ ∫ 𝑭(𝒙𝟏 , 𝒙𝟐 , 𝒚, 𝒚𝒙𝟏 , 𝒚𝒙𝟐 ) 𝒅𝒙𝟏 𝒅𝒙𝟐
𝑹
Solve a second-order partial differential equation:
𝝏 𝝏𝑭 𝝏 𝝏𝑭 𝝏𝑭
( )+ ( )− =𝟎
𝝏𝒙𝟏 𝝏𝒚𝒙𝟏 𝝏𝒙𝟐 𝝏𝒚𝒙𝟐 𝝏𝒚

5. Optimize:
𝒙𝟏
𝑰[𝒚(𝒙)] = ∫ 𝑭[𝒙, 𝒚, 𝒚′ ] 𝒅𝒙
𝒙𝟎
Subject to:
𝒙𝟏
𝑮(𝒙, 𝒚) = 𝟎 𝑱 = ∫ 𝑮(𝒙, 𝒚, 𝒚′ ) 𝒅𝒙 𝑮(𝒙, 𝒚, 𝒚′ ) = 𝟎
𝒙𝟎
𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠: 𝑎𝑙𝑔𝑒𝑏𝑟𝑎𝑖𝑐 𝑖𝑛𝑡𝑒𝑔𝑟𝑎𝑙 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡𝑖𝑎𝑙 𝑒𝑞𝑢𝑎𝑡𝑖𝑜𝑛

Form the Lagrangian function 𝐿 = 𝐹 + λ𝐺 and solve the Euler equation with the
constraint equation. The Lagrange Multiplier is a constant for integral constraints
and is a function of the independent variable for algebraic and differential
equation constraints.
__________________________________________________________________________
to the Euler-Poisson equation with higher order derivatives are available in
Weinstock (3).
When constraints are involved, the Lagrange function is formed, as
shown in the table. This gives an unconstrained problem that can be solved
by the Euler and/or Euler Poisson equation, along with the constraint
equations.
The purpose of the chapter was to give some of the key results of the
calculus of variations and to emphasize the similarities and differences
between finding an optimal function and an optimal point. Consequently, it
was necessary to select the methods given here from some equally important
methods that were omitted. Two of these are the concept of a variation and
the use of the second variation for the sufficient conditions to determine if
the function was actually a maximum or minimum. These are discussed by
Courant and Hilbert (5) along with the problem of the existence of a solution.
Also, most texts discuss the moving (or natural) boundary problem where
one or both of the limits on the integral to be optimized can be a function of
the independent variable. This leads to extensions of the Brachistochrone
problem, and Forray's discussion (1) is recommended. With the background
of this chapter, extension to Hamilton's principle follows, and is typically the
next topic presented on the subject. Also, this material leads to extensions
that include Pontryagin's maximum principle; Sturm-Liouville problems;
and application in optics, dynamics of particles, vibrations, elasticity, and
quantum mechanics.
The calculus of variations can be used to solve transport phenomena
problems, i.e., obtain solutions to the partial differential equations
representing the conservation of mass, momentum, and energy of a system.
In this approach the partial differential equations are converted to the
corresponding integral to be optimized from the calculus of variations. Then
approximate methods of integration are used to find the minimum of the
integral, and this yields the concentration, temperature, and/or velocity
profiles required for the solution of the original differential equations. This
approach is described by Schechter (4) in some detail.
Again, the purpose of the chapter was to introduce the topic of finding the
optimal function. The references at the end of the chapter are recommended
for further information; they include the texts by Fan (12) and Fan and Wang
(13) on the maximum principle and Kirk (14), among others, (7, 15, 16) on
optimal control.

REFERENCES

1. Forray, M. J., Variational Calculus in Science and Engineering, McGraw-Hill Book Co., New
York (1968).
 2. Ewing, G. M., Calculus of Variations with Applications, W. W. Norton Co., Inc., New York
(1969).
3. Weinstock, R., Calculus of Variations, McGraw-Hill Book Co., New York (1952).
4. Schechter, R. S., The Variational Method in Engineering, McGraw-Hill Book Co., New York
(1967).
5. Courant, R., and D. Hilbert, Methods of Mathematical Physics, Vol. 1, John Wiley & Sons,
Inc., New York (1953).
6. Sagan, H., Introduction to the Calculus of Variations, McGraw-Hill Book Co., New York
(1969).
7. M. M. Denn, Optimization by Variational Methods, McGraw-Hill Book Co., New York
(1970).
8. Wylie, C. R., and L. C. Barrett, Advanced Engineering Mathematics, 5th Ed., McGraw-Hill
Book Co., New York (1982).
9. Burley, D. M., Studies in Optimization, John Wiley & Sons, Inc., New York (1974).
10. Beveridge, G. S. G., and R. S. Schechter, Optimization: Theory and Practice, McGraw-Hill
Book Co., New York (1970).
11. Fan, L. T., E. S. Lee, and L. E. Erickson, Proc. of the Mid-Am. States Univ. Assoc. Conf on
Modern Optimization Techniques and their Application in Engineering Design, Part I, Kansas
State University, Manhattan, Kansas (Dec. 19-22, 1966).
12. L. T. Fan, The Continuous Maximum Principle, John Wiley & Sons Inc., New York (1966).
13. Fan, L. T. and C. S. Wang, The Discrete Maximum Principal, John Wiley & Sons, Inc., New
York (1964).
14. Kirk, D. E. Optimal Control Theory, An Introduction. Prentice-Hall, Inc., Englewood Cliffs,
N. J. (1970).
15. Miele, A., Optimization Techniques with Applications to Aerospace Systems, Ed., G. Leitman,
Ch.4, Academic Press, New York (1962).
16. Connors, M. M., and D. Teichroew, Optimal Control of Dynamic Operations Research
Models, International Textbook Co., Scranton, Pa. (1967).

PROBLEMS

8-1.16 A product is being produced at a steady rate of 𝑃0 𝑝𝑜𝑢𝑛𝑑𝑠 𝑝𝑒𝑟 ℎ𝑜𝑢𝑟.

It is necessary to change the production rate to 𝑃1 𝑝𝑜𝑢𝑛𝑑𝑠 𝑝𝑒𝑟 ℎ𝑜𝑢𝑟 and
minimize the cost resulting from raw material lost to off-specification
product and overtime wages during the transition period. This is modeled by
the following cost function:

𝑪(𝒕) = 𝒄𝟏 𝑷′𝟐 + 𝒄𝟐 𝒕𝑷′

where 𝑐1 and 𝑐2 are cost coefficients and 𝑃′ = 𝑑𝑃⁄𝑑𝑡. The total cost for the
change in the production schedule is given by:

𝒕𝟏
𝑪𝑻 = ∫ [𝒄𝟏 𝑷′𝟐 + 𝒄𝟐 𝒕𝑷′ ] 𝒅𝒕
𝒕𝟎

where 𝑃0 = 𝑃(𝑡0 ) and 𝑃1 = 𝑃(𝑡1 ) are known.

Determine the optimum way the production rate 𝑃(𝑡) is to be changed to
minimize the total cost.
8..2.15 A classical problem in aerodynamics is to determine the optimum
shape of a body of revolution that has the minimum drag. For a slender body
of revolution at zero angle of attack in an inviscid hypersonic flow, the total
drag is approximated by

𝑳
𝑫 = 𝟒𝝅𝝆𝝂𝟐 ∫ 𝒚(𝒚′ )𝟑 𝒅𝒙
𝟎

where 𝜈 and 𝜌 are the free stream velocity and density, respectively.
a. Obtain the differential equation and boundary conditions that are to
be solved to obtain the optimum body shape.
b. Show that the following is the solution to the differential equation
obtained in (a):

𝒚 = (𝒅⁄𝟐)(𝒙⁄𝑳)𝟑⁄𝟒

which, according to Miele (15), means that the contours of a body of

revolution having minimum drag for a given diameter, 𝑑, and a given
length 𝐿, is a parabola satisfying the 3⁄4-power law.
1
8-3. Determine the minimum surface of revolution by finding the curve
𝑦(𝑥) with prescribed end points such that by revolving this curve around the
𝑥 axis a surface of minimal area is obtained. The integral to be minimized is:

𝒙𝟏
𝑰 = 𝟐𝝅 ∫ 𝒚(𝟏 + 𝒚′𝟐 )𝟏⁄𝟐 𝒅𝒙
𝒙𝟎

a. Show that the Euler equation for 𝐹 = 𝐹(𝑦, 𝑦 ′ ) gives only:

𝝏𝑭
𝑭 − 𝒚′ = 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕
𝝏𝒚′

b. Apply this result to the problem to obtain:

𝒚(𝟏 + 𝒚′𝟐 )−𝟏⁄𝟐 = 𝒄𝟏

c. Define the parametric variable 𝑦 ′ = sinh 𝑡 in order to obtain a

more compact solution, and obtain the following result:

𝒙 = 𝒄𝟏 𝒕 + 𝒄𝟐

𝒚𝟏 = 𝒄𝟏 𝐜𝐨𝐬𝐡(𝒕)
 which is the parametric form of a family of catenaries, and 𝑐1 and
𝑐2 are boundary conditions for the end points of the curve.
9
8-4. Find the shape at equilibrium of a chain of length that hangs from
two points at the same level. The potential energy, 𝐸, of the chain is given
by the following equation:

𝒙𝟏
𝑬[𝒚(𝒙)] = −𝝆𝒈 ∫ 𝒚(𝟏 + 𝒚′𝟐 )𝟏⁄𝟐 𝒅𝒙
𝒙𝟎

and is subject to the specified total length 𝐿 by the following equation:

𝒙𝟏
𝑳 = ∫ (𝟏 + 𝒚′𝟐 )𝟏⁄𝟐 𝒅𝒙
𝒙𝟎

with boundary conditions of 𝑦(𝑥0 ) = 𝑦(𝑥1 ) = 0. To obtain the equilibrium

shape of the chain, it is necessary to minimize the energy subject to the length
restriction. Show that the following differential equation is obtained from the
Euler equation.

𝒚′ = [𝒌𝟐 (𝒚 + 𝝀)𝟐 − 𝟏]𝟏⁄𝟐

Make the substitution 𝑘(𝑦 + 𝜆) = cosh 𝜃, and obtain the solution given
below:

𝐜𝐨𝐬𝐡 𝒌(𝒙 + 𝜶) = 𝒌(𝒚 + 𝝀)

This curve is the catenary, and the constants 𝑘, 𝛼, 𝜆 can be obtained from the
boundary conditions and the constraint on 𝐿.
8-5.14 A simple optimal control problem related to an electromechanical
system can be formulated as:

𝑻
𝑴𝒊𝒏𝒊𝒎𝒊𝒛𝒆: 𝑰 = ∫ [𝒚𝟐 𝟐 (𝒕) − 𝒚𝟑 𝟐 (𝒕)] 𝒅𝒕
𝟎

𝑺𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐: 𝒚𝟏 ′ + 𝒚𝟏 = 𝒚𝟑

𝒚 𝟐 ′ − 𝒚𝟏 = 𝟎

a. Obtain the differential equations to be solved for the optimal

functions. Show that there are sufficient equations to determine the
dependent variables.
b. What boundary conditions are required?
8-6.9 For steady flow of an incompressible fluid in a square duct, the
equations of continuity and motion simplify to a partial differential equation
that requires an elaborate analytical solution involving the sum of an infinite
series. An approximate solution can be obtained using the calculus of
variations, which gives a simple equation that predicts the volumetric flow
rate within 1% of the exact solution. The equation that describes the flow at
a point 𝑧 along the axis of the duct is:

𝝏𝟐 𝝂 𝝏𝟐 𝝂 𝟏 𝒅𝑷
+ =
𝝏𝒙𝟐 𝝏𝒚𝟐 𝝁 𝒅𝒛

where 𝜈 is the axial velocity; 𝜇 is the viscosity of the fluid; 𝑑𝑃⁄𝑑𝑧 is the
pressure gradient, a constant; and 𝑎 is the length of one-half of the side of
the duct. The boundary conditions are that there is no slip at the wall, i.e.,
𝜈 = 0 at 𝑥 = 𝑎 for 0 < 𝑦 < 𝑎 and at 𝑦 = 𝑎 for 0 < 𝑥 < 𝑎.
a. Show that the integral to be minimized corresponding to the
differential equation is:

𝒂 𝒂
𝝏𝝂 𝟐 𝝏𝝂 𝟐 𝟐 𝒅𝑷
𝑰[𝝂(𝒙, 𝒚)] = ∫ ∫ [( ) + ( ) − 𝝂] 𝒅𝒙𝒅𝒚
𝟎 𝟎 𝝏𝒙 𝝏𝒚 𝝁 𝒅𝒛

b. A simple approximation to 𝜈 is given by the following equation,

which satisfies the boundary conditions:

𝝂 = 𝑨(𝒂𝟐 − 𝒙𝟐 )(𝒂𝟐 − 𝒚𝟐 )

Using this equation, perform the integration of the equation in (a) to

obtain the following result:

𝑰[𝝂] = (𝟔𝟓⁄𝟒𝟓)𝑨𝟐 𝒂𝟖 − (𝟖⁄𝟗𝝁)(𝒅𝑷⁄𝒅𝒛)𝑨𝒂𝟔

c. Find the value of 𝐴 that minimizes 𝐼.

d. If the mass flow rate, 𝑤, through the duct is given by the following.
equation:
𝒂 𝒂
𝒘 = 𝝆 ∫ ∫ 𝝂𝒅𝒙𝒅𝒚
𝟎 𝟎

show that the following result is obtained:

𝒘 = 𝟎. 𝟓𝟓𝟔𝝆(𝒅𝑷⁄𝒅𝒛)𝒂𝟒 ⁄𝝁
 The analytical solution has the same form, but the coefficient is
0.560. Thus, the approximate solution is within 1% of the exact
solution.
8-7. In a production scheduling problem, the production rate is to be changed
from 100 units per unit time to 300 units per unit time in 10 time units, i.e.,
𝑝(0) = 100 and 𝑝(10) = 300. The costs as a function of time are associated
with changes in machines, personnel, and raw materials. For this simple
problem this cost is given as:

𝒄(𝒕) = 𝟐(𝒑′ ) + 𝟒𝒕𝒑′

where 𝑝′ = 𝑑𝑝⁄𝑑𝑡
Determine the production rate as a function of time that minimizes the
cost over the time.
8-8.1 Determine the deflection in an uniformly loaded, cantilever beam,
𝑦(𝑥), where 𝑦 is the deflection as a function of distance down the beam from
the wall (𝑥 = 0) to the end of the beam (𝑥 = 𝐿). The total potential energy
of the system to be minimized is given by:

𝑳
𝑰[𝒚(𝒙)] = ∫ [(𝑬⁄𝟐)(𝒚′′ )𝟐 − 𝒒𝒚] 𝒅𝒙
𝟎

where 𝐸 is the bending rigidity and 𝑞 is the load. The boundary conditions
at the wall end are 𝑦(0) = 𝑦 ′ (0) = 0, and at the supported end are 𝑦 ′′′ (𝐿) =
𝑦 ′′ (𝐿) = 0.