CHAPTER 6

Some Methods in the

Calculus of Variations

6-1. If we use the varied function

y (α , x ) = x + α sin π (1 − x ) (1)

Then
dy
= 1 − απ cos π (1 − x ) (2)
dx
Thus, the total length of the path is
1 2
 dy 
S=∫ 1 +   dx
0
 dx 

1
= ∫  2 − 2απ cos π (1 − x ) + α 2π 2 cos 2 π (1 − x ) 
12
dx (3)
0

Setting π (1 − x ) ≡ u , the expression for S becomes

π 12
1  1 
S=
π ∫ 2 1 − απ cos u + α 2π 2 cos 2
 2
u

du (4)
0

The integral cannot be performed directly since it is, in fact, an elliptic integral. Because α is a
small quantity, we can expand the integrand and obtain
π
2  1 1 2 2 2  1 1 2 2 2 
2

S=
π ∫ 1 −  απ cos u − α π cos u −  απ cos u − α π cos u + … du (5)
0  2 2 8 2 
If we keep the terms up to cos 2 u and perform the integration, we find

2 2 2
S= 2+ π α (6)
16
which gives

165
166 CHAPTER 6

∂S 2 2
= π α (7)
∂α 8
Therefore

∂S
=0 (8)
∂α a=0

and S is a minimum when α = 0.

6-2. The element of length on a plane is

dS = dx 2 + dy 2 (1)

from which the total length is

( x2 , y2 ) x2 2
 dy 
S= ∫ dx 2 + dy 2 = ∫ 1 +   dx
 dx 
(2)
( x1 , y1 ) x1

If S is to be minimum, f is identified as
2
 dy 
f = 1+   (3)
 dx 

Then, the Euler equation becomes

d d 
1 + y′2  = 0 (4)
dx dy ′  

dy
where y ′ = . (4) becomes
dx

d  y′ 
 =0 (5)
dx  1 + y ′ 2 

or,
y′
= constant ≡ C (6)
1 + y′2

from which we have

C2
y′ = = constant ≡ a (7)
1 − C2
Then,

y = ax + b (8)

This is the equation of a straight line.

SOME METHODS IN THE CALCULUS OF VARIATIONS 167

6-3. The element of distance in three-dimensional space is

dS = dx 2 + dy 2 + dz 2 (1)

Suppose x, y, z depends on the parameter t and that the end points are expressed by
( x1 (t1 ) , y1 (t1 ) , z1 (t1 )) , ( x2 (t2 ) , y2 (t2 ) , z2 (t2 )) . Then the total distance is
t2 2 2 2
 dx   dy   dz 
S=∫  dt  +  dt  +  dt  dt (2)
t1

The function f is identified as

f = x2 + y2 + z2 (3)

∂ f ∂f ∂ f
Since = = = 0 , the Euler equations become
∂x ∂y ∂z

d ∂f 
=0 
dt ∂x 

d ∂f
=0  (4)
dt ∂y 

d ∂f 
=0 
dt ∂z 
from which we have

x 
= constant ≡ C1 
x2 + y2 + z2 

y 
= constant ≡ C2  (5)
x2 + y2 + z2 

z 
= constant ≡ C3 
x2 + y2 + z2 

From the combination of these equations, we have
x y 
=
C1 C2 

 (6)
y z 
=
C2 C3 

If we integrate (6) from t1 to the arbitrary t, we have

168 CHAPTER 6

x − x1 y − y1 
=
C1 C2 

 (7)
y − y1 z − z1 
=
C2 C3 

On the other hand, the integration of (6) from t1 to t2 gives

x2 − x1 y2 − y1 
=
C1 C2 

 (8)
y2 − y1 z2 − z1 
=
C2 C3 

from which we find the constants C1 , C2 , and C3 . Substituting these constants into (7), we find

x − x1 y − y1 z − z1
= = (9)
x2 − x1 y2 − y1 z2 − z1

This is the equation expressing a straight line in three-dimensional space passing through the
two points ( x1 , y1 , z1 ) , ( x2 , y2 , z2 ) .

6-4.
z

1
ρ
dS
2
y
x
φ

The element of distance along the surface is

dS = dx 2 + dy 2 + dz 2 (1)

In cylindrical coordinates (x,y,z) are related to (ρ,φ,z) by

x = ρ cos φ 

y = ρ sin φ  (2)


z=z 
from which
dx = − ρ sin φ dφ 

dy = ρ cos φ dφ  (3)


dz = dz 
SOME METHODS IN THE CALCULUS OF VARIATIONS 169

Substituting (3) into (1) and integrating along the entire path, we find
2 φ2
S = ∫ ρ dφ + dz =
2 2 2
∫ ρ 2 + z 2 dφ (4)
1 φ1

dz
where z = . If S is to be minimum, f ≡ ρ 2 + z 2 must satisfy the Euler equation:
dφ
∂f ∂ ∂f
− =0 (5)
∂ z ∂φ ∂ z

∂f
Since = 0 , the Euler equation becomes
∂z
∂ z
=0 (6)
∂φ ρ + z2
2

from which
z
= constant ≡ C (7)
ρ + z2
2

or,

C2
z= ρ (8)
1 − C2
Since ρ is constant, (8) means

dz
= constant
dφ

and for any point along the path, z and φ change at the same rate. The curve described by this
condition is a helix.

6-5.
z
(x2,y2)
ds
(x1,y1)
y
x

The area of a strip of a surface of revolution is

dA = 2π × ds = 2π × dx 2 + dy 2 (1)

Thus, the total area is

170 CHAPTER 6

x2

A = 2π ∫x 1 + y 2 dx (2)
x1

dy
where y = . In order to make A a minimum, f ≡ x 1 + y 2 must satisfy equation (6.39). Now
dx
∂f
= 1 + y2
∂x
∂f xy
=
∂y 1 + y2

Substituting into equation (6.39) gives

d  xy 2  d  x 
1 + y2 = x 1 + y2 − =  
dx  1 + y2  dx  1 + y 2 


1 + y 2 − xy ( dy dx ) 1 + y 2 ( )
−1 2

=
1 + y2

Multiplying by 1 + y 2 and rearranging gives

dx dy
− =
x y 1 + y2 ( )
Integration gives

1 y2
− ln x + ln a = ln
2 1 + y2
where ln a is a constant of integration. Rearranging gives
1
y2 =
( x a2 − 1
2
)
Integrating gives
x
y = b + a cosh −1
a
or
y−b
x = a cosh
a
which is the equation of a catenary.
SOME METHODS IN THE CALCULUS OF VARIATIONS 171

6-6.
0 θ=π
y

(x1,y1)
θ=0
2a
(x2,y2)
x

If we use coordinates with the same orientation as in Example 6.2 and if we place the minimum
point of the cycloid at (2a,0) the parametric equations are

x = a (1 + cos θ ) 
 (1)

y = a (θ + sin θ ) 

Since the particle starts from rest at the point ( x1 , y1 ) , the velocity at any elevation x is [cf. Eq.
6.19]

v = 2 g ( x − x1 ) (2)

Then, the time required to reach the point ( x2 , y2 ) is [cf. Eq. 6.20]
12
x2
 1 + y′2 
t= ∫   dx (3)
x1  2 g ( x − x1 ) 
Using (1) and the derivatives obtained therefrom, (3) can be written as
θ1 12
a  1 + cos θ 
t=
g ∫  cos θ − cosθ1  dθ (4)
θ2 = 0

Now, using the trigonometric identity, 1 + cos θ = 2 cos 2 θ 2 , we have

θ
θ1 cos dθ
a
t=
g ∫ θ
2
θ1
0
cos 2 − cos 2
2 2
θ
θ1 cos dθ
a
=
g ∫ θ1
2
θ
(5)
0
sin 2
− sin 2

2 2
Making the change of variable, z = sin θ 2 , the expression for t becomes
θ1
sin
2
a dz
t=2
g ∫ θ1
(6)
0
sin 2 − z2
2
The integral is now in standard form:
172 CHAPTER 6

dx x
∫ a =x
2
= sin −1  
2
a
(7)

Evaluating, we find

a
t=π (8)
g

Thus, the time of transit from ( x1 , y1 ) to the minimum point does not depend on the position of
the starting point.

6-7.

a θ1 θ1
n1
(n2 > n1)
n2
c
b θ2 v1 =
n1
c
v2 =
x n2

The time to travel the path shown is (cf. Example 6.2)

ds 1 + y′2
t=∫ =∫ dx (1)
v v
Although we have v = v(y), we only have dv dy ≠ 0 when y = 0. The Euler equation tells us

d  y′ 
 =0 (2)
dx  v 1 + y ′ 2
 
Now use v = c n and y′ = –tan θ to obtain

n sin θ = const. (3)

This proves the assertion. Alternatively, Fermat’s principle can be proven by the method
introduced in the solution of Problem 6-8.

6-8. To find the extremum of the following integral (cf. Equation 6.1)

J = ∫ f ( y , x ) dx

we know that we must have from Euler’s equation

∂f
=0
∂y
This implies that we also have
SOME METHODS IN THE CALCULUS OF VARIATIONS 173

∂J ∂f
=∫ dx = 0
∂y ∂y
giving us a modified form of Euler’s equation. This may be extended to several variables and to
include the imposition of auxiliary conditions similar to the derivation in Sections 6.5 and 6.6.
The result is

∂J ∂g j
+ ∑ λ j ( x) =0
∂y i j ∂y i

when there are constraint equations of the form

g j ( yi , x ) = 0

a) The volume of a parallelepiped with sides of lengths a1 , b1 , c1 is given by

V = a1b1c1 (1)
We wish to maximize such a volume under the condition that the parallelepiped is
circumscribed by a sphere of radius R; that is,
a12 + b12 + c12 = 4 R 2 (2)

We consider a1 , b1 , c1 as variables and V is the function that we want to maximize; (2) is the
constraint condition:
g { a1 , b1 , c1 } = 0 (3)

Then, the equations for the solution are

∂V ∂g 
+λ =0 
∂a1 ∂a1 

∂V ∂g 
+λ =0  (4)
∂b1 ∂b1


∂V ∂g
+λ =0 
∂c1 ∂c1 
from which we obtain
b1c1 + 2λ a1 = 0 

a1c1 + 2λ b1 = 0  (5)


a1b1 + 2λ c1 = 0 

Together with (2), these equations yield

2
a1 = b1 = c1 = R (6)
3
2
Thus, the inscribed parallelepiped is a cube with side R.
3
174 CHAPTER 6

b) In the same way, if the parallelepiped is now circumscribed by an ellipsoid with semiaxes
a, b, c, the constraint condition is given by
a12 b12 c12
= = =1 (7)
4 a 2 4b 2 4 z 2
where a1 , b1 , c1 are the lengths of the sides of the parallelepiped. Combining (7) with (1) and
(4) gives
a12 b12 c12
= = (8)
a2 b 2 c2
Then,

2 2 2
a1 = a , b1 = b , c1 = c (9)
3 3 3

6-9. The average value of the square of the gradient of φ ( x1 , x2 , x3 ) within a certain volume V
is expressed as
1
∫∫∫ ( ∇φ )
2
I= dx1 dx2 dx3
V

1   ∂φ  2  ∂ φ  2  ∂ φ  2 
v= ∫∫∫ 
∂x
 +
∂x
 +
∂x
  dx1dx2 dx3 (1)
  1   2   3  
V

In order to make I a minimum,

2 2 2
 ∂φ   ∂φ   ∂φ 
f =  +  + 
 ∂x1   ∂x2   ∂x3 
must satisfy the Euler equation:
∂f 3
∂  ∂f 
−∑ =0 (2)
∂ φ i = 1 ∂x i   ∂ φ  
∂  
  ∂xi  

If we substitute f into (2), we have

3
∂ ∂φ
∑ ∂x =0 (3)
i =1 i ∂x i

which is just Laplace’sequation:

∇2 φ = 0 (4)

Therefore, φ must satisfy Laplace’s equation in order that I have a minimum value.
SOME METHODS IN THE CALCULUS OF VARIATIONS 175

6-10. This problem lends itself to the method of solution suggested in the solution of Problem
6-8. The volume of a right cylinder is given by
V = π R2 H (1)
The total surface area A of the cylinder is given by
A = Abases + Aside = 2π R 2 + 2π RH = 2π R ( R + H ) (2)
We wish A to be a minimum. (1) is the constraint condition, and the other equations are
∂A ∂g 
+λ =0 
∂R ∂R
 (3)
∂A ∂g 
+λ =0 
∂H ∂H 
where g = V − π R 2 H = 0 .

The solution of these equations is

1
R= H (4)
2

6-11.
y

1 θ
R ds
2a
}

The constraint condition can be found from the relation ds = Rdθ (see the diagram), where ds is
the differential arc length of the path:

( )
12
ds = dx 2 + dy 2 = Rdθ (1)

which, using y = ax 2 , yields

1 + 4a 2 x 2 dx = Rdθ (2)

If we want the equation of constraint in other than a differential form, (2) can be integrated to
yield

A + Rθ =
x
2
4 ax 2 + 1 +
1
4a
(
ln 2ax + 4 a 2 x 2 + 1 ) (3)

where A is a constant obtained from the initial conditions. The radius of curvature of a parabola,
y = ax 2 , is given at any point (x,y) by r0 ≥ 1 2a . The condition for the disk to roll with one and
only one point of contact with the parabola is R < r0 ; that is,
176 CHAPTER 6

1
R< (4)
2a

6-12. The path length is given by

s = ∫ ds = ∫ 1 + y ′ 2 + z ′ 2 dx (1)

and our equation of constraint is

g ( x, y, z) = x2 + y 2 + z2 − ρ2 = 0 (2)

The Euler equations with undetermined multipliers (6.69) tell us that

d  y′  dg
 =λ = 2λ y (3)
dx  1 + y ′ 2 + z ′ 2  dy

with a similar equation for z. Eliminating the factor λ, we obtain

1 d  y′  1 d  z′ 
 −  =0 (4)
y dx  1 + y ′ 2 + z ′ 2  z dx  1 + y ′ 2 + z ′ 2 

This simplifies to

( ) ( )
z  y ′′ 1 + y ′ 2 + z ′ 2 − y ′ ( y ′y ′′ + z ′z ′′ )  − y  z ′′ 1 + y ′ 2 + z ′ 2 − z ′ ( y ′y ′′ + z ′z ′′ )  = 0 (5)

zy ′′ + ( yy ′ + zz ′ ) z ′y ′′ − yz ′′ − ( yy ′ + zz ′ ) y ′z ′′ = 0 (6)

and using the derivative of (2),

( z − xz ′ ) y ′′ = ( y − xy ′ ) z ′′ (7)

This looks to be in the simplest form we can make it, but is it a plane? Take the equation of a
plane passing through the origin:
Ax + By = z (8)

and make it a differential equation by taking derivatives (giving A + By′ = z′ and By″ = z″) and
eliminating the constants. The substitution yields (7) exactly. This confirms that the path must
be the intersection of the sphere with a plane passing through the origin, as required.

6-13. For the reason of convenience, without lost of generality, suppose that the closed curve
passes through fixed points A(-a,0) and B(a,0) (which have been chosen to be on axis Ox). We
denote the part of the closed curve above and below the Ox axis as y1 ( x ) and y2 ( x ) respectively.
(note that y1 > 0 and y2 < 0 )

The enclosed area is

a a a a
J ( y1 , y2 ) = ∫ y1 ( x)dx − ∫ y2 ( x)dx = ∫ ( y1 (x) − y2 (x)) dx = ∫ f ( y1 , y2 )dx
−a −a −a −a
SOME METHODS IN THE CALCULUS OF VARIATIONS 177

The total length of closed curve is

∫{ }
a a a a
K ( y1′ , y2′ ) = ∫ 1 + ( y1′ ) dx + ∫ 1 + ( y2′ ) dx = 1 + ( y1′ ) + 1 + ( y1′ ) dx = ∫ g ( y1′ , y2′ ) dx
2 2 2 2

−a −a −a −a

Then the generalized versions of Eq. (6.78) (see textbook) for this case are

∂f d ∂f  ∂g d ∂g  d  y1′ 
− +λ −  = 0 ⇒ 1− λ  =0 (1)
∂y1 dx ∂y1′  ∂y1 dx ∂y1′  dx  1 + ( y1′ ) 
2

∂f d ∂f  ∂g d ∂g  d  y2′ 
− +λ −  = 0 ⇒ 1− λ  =0 (2)
∂y2 dx ∂y2′  ∂y2 dx ∂y2′  dx  1 + ( y2′ ) 
2

Analogously to Eq. (6.85);

from (1) we obtain ( x − A1 ) 2 + ( y1 − A2 ) 2 = λ 2 (3)

from (2) we obtain ( x − B1 ) 2 + ( y2 − B2 ) 2 = λ 2 (4)

where constants A’s, B’s can be determined from 4 initial conditions

( x = ± a, y1 = 0) and ( x = ± a, y2 = 0)
We note that y1 < 0 and y2 > 0 , so actually (3) and (4) altogether describe a circular path of
radius λ . And this is the sought configuration that renders maximum enclosed area for a given
path length.

6-14. It is more convenient to work with cylindrical coordinates (r, φ ,z) in this problem. The
constraint here is z = 1 – r , then dz = –dr

ds2 = dr 2 + r 2 dφ 2 + dz 2 = 2 ( dr 2 + r 2 dβ 2 )

φ
where we have introduced a new angular coordinate β =
2
In this form of ds2 , we clearly see that the space is 2-dimensional Euclidean flat, so the shortest
line connecting two given points is a straight line given by:
r0 r0
r= =
cos ( β − β 0 )  φ − φ0 
cos 
 2 

π
this line passes through the endpoints (r = 1, φ = ± ) , then we can determine unambiguously
2
the shortest path equation
π
cos
r(φ ) = 2 2 and z = 1 – r
 φ 
cos 
 2 
178 CHAPTER 6

Accordingly, the shortest connecting length is

π 2
π
2
 dr 
l= ∫ dφ 2  dφ  + r = 2 2 sin 2 2
2

−π 2

6-15.
1
 dy  2 
I[ y] = ∫   − y 2  dx
 
0  dx 

a) Treating I[y] as a mechanical action, we find the corresponding Euler-Lagrange equation

d2 y
y( x ) = −
dx 2
Combining with the boundary conditions (x = 0, y = 0) and (x = 1, y = 1), we can determine
unambiguously the functional form of y( x ) = (sin x) (sin 1) .

b) The corresponding minimum value of the integral is

1
 dy  2  1
1
I[ y] = ∫   − y 2  dx =
sin 2 1 ∫0
dx cos 2 x = cot (1) = 0.642
 
0 dx 

c) If x = y then I[y] = ( 2 3 ) = 0.667.

6-16.
a) S is arc length
2 2 2
 dy   dz   dy  9
S = ∫ dx 2 + dy 2 + dz 2 = ∫ dx 1 +   +   = ∫ dx 1 +   + x = ∫ L dx
 dx   dx   dx  4
Treating S and L like a mechanical action and Lagrangian respectively, we find the canonical
momentum associated with coordinate y
 dy 
δL  
p= = dx
 dy  9  dy 
2
δ  1+ x+ 
 dx  4  dx 

Because L does not depend on y explicitly, then E-L equation implies that p is constant
(i.e. dp dx = 0 ), then the above equation becomes

9
1+ x 32
dy 4 ⇒y = p 9  9 
1 − p2 ∫
=p dx 1 + x = A 1 + x +B
dx 1 − p2 4  4 

where A and B are constants. Using boundary conditions (x = 0, y = 0) and (x = 1, y = 1) one can
determine the arc equation unambiguously
SOME METHODS IN THE CALCULUS OF VARIATIONS 179

 
32
8 9 
y( x ) = 3 2  1 + x − 1 and z = x3 2
13 − 8  4 
b)
z

0.75

0.5 x

0.25 1
y 0.
0.75
0
1 0.5
0.75 0.25
0.5
0.25
00

6-17.
a) Equation of a ellipse
x2 y2
+ =1
a2 b2
which implies

ab 2xy x 2 y 2
xy ≤ because ≤ 2 + 2
2 ab a b
so the maximal area of the rectangle, whose corners lie on that ellipse, is
Max[A] = Max[4xy] = 2 iab.
This happens when
a b
x= and y=
2 2
b) The area of the ellipse is A0 = π ab ; so the fraction of rectangle area to ellipse area is then
Max[ A] 2
=
A0 π

6-18. One can see that the surface xy = z is “locally” symmetric with respect to the line
x = − y = −z where x > 0, y < 0, z < 0. This line is a parabola. This implies that if the particle
starts from point (1,-1,-1) (which belongs to the symmetry line) under gravity ideally will move
downward along this line. Its velocity at altitude z (z < –1) can be found from the conservation
of energy.

v( z) = −2 g( z + 1)
180 CHAPTER 6