Faculty of Science, Technology, Engineering and Mathematics

Specimen Solutions
M820 Calculus of variations and advanced calculus

M820 SPECIMEN SOLUTIONS 2 2021J

TMA 02 Cut-off date 2 February 2022

Solution to Question 1
2y 02
(a) Here, F = y 2 sinh x − and so the Euler-Lagrange equation is
sinh x
4y 0
 
d
− − 2y sinh x = 0,
dx sinh x
which simplifies to
2y 00 sinh x − 2y 0 cosh x
+ y sinh x = 0,
sinh2 x
so the Euler-Lagrange equation is

2y 00 sinh x − 2y 0 cosh x + y sinh3 x = 0, y(a) = A, y(b) = B.

(b) If x is a function of the new variable u then the functional becomes

Z u2 "  2 #
dx 2 2 dy du
S[y] = du y sinh x − ,
u1 du sinh x du dx

where x(u1 ) = a and x(u2 ) = b.

If (dx/du) sinh x = 1 then the functional becomes
Z u2  2 !
dy
du y 2 − 2 ,
u1 du

as required.
Integrating (dx/du) sinh x = 1 gives u = cosh x, and the required limits are
therefore u1 = cosh a and u2 = cosh b.
(c) The Euler-Lagrange equation for the new functional is
d
(−4y 0 (u)) − 2y = 0, ie 2y 00 (u) + y = 0.
du
This has general solution
√ √
y = C cos(u/ 2) + D sin(u/ 2),
where C and D are constants.

Copyright c 2021 The Open University

(d) From above, the general solution of 2y 00 sinh x − 2 cosh xy 0 + y sinh3 x = 0 is
1 1
y = C cos( √ cosh x) + D sin( √ cosh x),
2 2
which can also be written as
1
y = E sin( √ cosh x + F ).
2
This latter form is better when applying the boundary conditions; y(1) = 0
immediately gives
1
y = E sin( √ (cosh x − cosh 1)).
2
Then applying y(2) = 2 gives
2
E= ,
sin( √12 (cosh 2 − cosh 1))

so that
2 sin( √12 (cosh x − cosh 1))
y=
sin( √12 (cosh 2 − cosh 1))
as required.

Solution to Question 2
(a) The Euler-Lagrange equations for y1 and y2 are, respectively,
d d
(2y 0 ) − 4(2y1 + y2 ) = 0 and (4y 0 ) − 2(2y1 + y2 ) = 0.
dx 1 dx 2
These expand to
y100 − 4y1 − 2y2 = 0 and 2y200 − 2y1 − y2 = 0.

(b) If z1 = y1 + ay2 and z2 = 2y1 + y2 then

az2 − z1 2z1 − z2
y1 = and y2 = ,
2a − 1 2a − 1
where we assume 2a − 1 6= 0, otherwise z1 and z2 are not independent.
Then the functional becomes
Z h i
 0
S[z1 , z2 ] = 1
dx (2a−1) 2 (az2 − z10 )2 + 2(2z10 − z20 )2 + z22
Z h i
1
9z102 + (2 + a2 )z202 − 2(a + 4)z10 z20 + z22 .

= dx (2a−1)2

The transformation gives a separable functional if the coefficient of z10 z20 is zero,
that is if a = −4.
(c) For a = −4 the separable functional becomes
Z
1
S[z1 , z2 ] = dx 81 [9z102 + 18z202 + 81z22 ],

and the resulting uncoupled Euler-Lagrange equations are

d d
(18z10 ) = 0 and (36z20 ) − 162z2 = 0,
dx dx
that is
z100 = 0 and z200 − 29 z2 = 0.

page 2 of ??
 The general solution of these equations is

z1 = Ax + B, z2 = Cemx + De−mx ,
√
where m = 3/( 2) and A, B, C and D are constants.
Since y1 = − 19 (−4z2 − z1 ) and y2 = − 19 (2z1 − z2 ), we have the general solution
for (y1 , y2 ) as
y1 = −4(Cemx + De−mx ) − (Ax + B), y2 = 2(Ax + B) − (Cemx + De−mx ),
where the factor of −1/9 has been absorbed into the A, B, C and D.
(d) For y1 and y2 as given above,
y100 − 4y1 − 2y2 = −4(m2 Cemx + m2 De−mx ) + 16(Cemx + De−mx )
+4(Ax + B) − 4(Ax + B) + 2(Cemx + De−mx )
= (−4m2 + 18)Cemx + (−4m2 + 18)De−mx = 0
and
2y200 − 2y1 − y2 = −2(m2 C mx + m2 De−mx ) + 8(Cemx + De−mx )
+2(Ax + B) − 2(Ax + B) + (Cemx + De−mx )
= (−2m2 + 9)Cemx + (−2m2 + 9)D−mx = 0,
as required, since m2 = 9/2.

Solution to Question 3
(a) Writing
 
0 5 0 2 2 3
F (x, y, y ) = x (y ) − y ,
3
the Euler-Lagrange equation is
 
d ∂F ∂F
− =0
dx ∂y 0 ∂y
d
2x5 y 0 − −2x5 y 2 = 0



dx
which simplifies to
d2 y 5 dy
+ + y2 = 0 ,
dx2 x dx
as required.
(b) Let x̄ = αx, ȳ = βy, where α, β constants.
Then, denoting ā = αa, b̄ = αb, we have
Z b̄  
5 0 2 2 3
S̄[ȳ] = dx̄ x̄ (ȳ ) − ȳ
ā 3
Z b  2 !
5 5 β 0 2 3 3
= αdx α x y − β y
a α 3
Z b  
2
= dx x5 (α2 β)2 (y 0 )2 − (α2 β)3 y 3
a 3
= S[y] ,
provided α2 β = 1.

page 3 of ??
(c) We now apply Noether’s Theorem. Consider the one-parameter family with
α=1+δ
x̄ = (1 + δ)x = x + δφ , φ=x
−2
ȳ = (1 + δ) y = y + δψ + O(δ 2 ) , ψ = −2y ,
where δ is the parameter.
Then, by Noether’s theorem, we have a first-integral given by
 
∂F 0 ∂F
ψ+ F −y φ = const.
∂y 0 ∂y 0
   
2
x5 2y 0 (−2y) + x5 (y 0 )2 − y 3 − 2x5 (y 0 )2 x = const.
3
 
5 0 6 0 2 2 3
4x yy + x (y ) + y = c ,
3
where c is a constant.
(d) Let
B
y1 = Ax−2 , y2 = .
(x2 + B/24)2
For brevity, let us write z = x2 + B/24.
Then we have
−4Bx
y10 = −2Ax−3 , y20 =
z3
−4B 24Bx2
y100 = 6Ax−4 , y200 = 3 + .
z z4
Thus, for y = y1 :
   
0 0 2 2 3 −2 −3 2 −6 2
5 6
4x yy + x (y ) + y = 5
4x Ax (−2Ax )+x 6
4A x + A3 x−6
3 3
2 3
A − 4A2 ,
=
3
which a constant, so that the first-integral is satisfied.
Now
5 0 5
y 00 + y + y2 = 6Ax−4 + (−2Ax−3 ) + A2 x−4
x x

= x−4 A2 − 4A .
Since this is zero iff A = 0 or A = 4, y1 is not a solution of the Euler-Lagrange
equation for arbitrary A.
Now for y = y2 :
24Bx2 B2
 
5 −4B 5 −4Bx
y 00 + y 0 + y 2 = + + +
x z3 z4 x z3 z4
1
−4B(x2 + B/24) + 24Bx2 − 20B(x2 + B/24) + B 2


=
z4
= 0,
so that y2 satisfies the Euler-Lagrange equation for all B.
To show that for y2 the first-integral holds, we can either appeal to Noether’s
theorem, or show directly as follows:
16B 2 x2 2 B3
      
2 B −4Bx
4x5 yy 0 + x6 (y 0 )2 + y 3 = 4x5 + x 6
+
3 z2 z3 z6 3 z6
2x6 B 2
−24(x2 + B/24) + 24x2 + B


= 6
3z
= 0,
a constant, as required.
Note: We should exclude those B < 0 for which y2 (x) has a singularity in the
interval [a, b], although the Euler-Lagrange equation and the first integral are
satisfied at non-singular points.

page 4 of ??
Solution to Question 4
(a)
Z 2
S[y] = dx(y 0 )n ey , y(0) = 1, y(1) = A > 1
0

The first-integral is
y 0 n(y 0 )n−1 ey − (y 0 )n ey = (n − 1)c̃n , c̃ constant
so

(y 0 )n = c̃e−y
or
y 0 = ĉe−y/n ,

where ĉn = c̃.

This equation has general solution
y = n ln(cx + α)
where c = ĉ/n and α is a constant of integration.
From the boundary conditions we have
y(0) = 1, so
e1/n = α, and
y(1) = A, so
c = eA/n − e1/n

and y = n ln(cx + e1/n ) where c = eA/n − e1/n , as required.

We note that c > 0 since A > 1.
(b) The Jacobi equation is
 
d du
P − Qu = 0, u(0) = 0, u0 (0) = 1
dx dx
where
∂2F
P =
∂y 02
∂2F
 2 
d ∂ F
Q= −
∂y 2 dx ∂y∂y 0
and F (y, y 0 ) = (y 0 )n ey
Then
 n−2
0 n−2 y nc
P = n(n − 1)(y ) e = n(n − 1) (cx + e1/n )n
cx + e1/n
= n(n − 1)(cn)n−2 (cx + e1/n )2
∂2F
= n(y 0 )n−1 ey
∂y∂y 0
 2 
d ∂ F d
n(y 0 )n−1 ey


0
=
dx ∂y∂y dx
 n−1 !
d cn 1/n n
= n (cx + e )
dx (cx + e1/n )
= n(cn)n−1 c = (cn)n
∂2F
 
0 n y nc n
= (y ) e = (cx + e1/n )n = (cn)n
∂y 2 cx + e1/n

page 5 of ??
 so
Q = (cn)n − (cn)n = 0
and the Jacobi equation is
 
d du
P = 0.
dx dx
This can be integrated once to get
du β
=
dx (cx + e1/n )2
for some constant β.
Thus
γ
u= +δ
(cx + e1/n )
for constants γ and δ.
From the initial conditions u(0) = 0, u0 (0) = 1 we obtain the solution

e1/n e2/n
u= − .
c c(cx + e1/n )

Now, since c > 0, u > 0 for all x > 0 so there are no conjugate points
on (0, ∞).
The nature of the stationary path is determined by the sign of
P = n(n − 1)(cn)n−2 (cx + e1/n )2 . This is positive for n > 1, so in this case the
stationary path is a local weak minimum.

Solution to Question 5
We have a < b, A, B > 0, f (x) > 0 on [a, b], and
Z b
S[y] = dx f (x)(1 + y 02 )1/2 , y(a) = A , y(b) = B .
a

(a) The Euler–Lagrange equation is

y0
 
d
f (x) = 0,
dx (1 + y 02 )1/2
so that
y0
f (x) =β, where β is constant. (1)
(1 + y 02 )1/2
Solving for y 0 gives
f (x)2 y 02 = β 2 1 + y 02 ,

so
β2
y 02 =
f (x)2 − β 2
and hence
β
y0 =
(f (x)2 − β 2 )1/2
since y 0 and β have the same signs.
Integrating gives
Z x
1
y =α+β dw p .
a f (w)2 − β 2

page 6 of ??
 Using the boundary conditions give

A = y(a) = α
Z b
1
B = y(b) = A + β dw p ,
a f (w)2 − β 2
as required.
(b) Denoting F (x, y 0 ) = f (x)(1 + y 02 )1/2 , we calculate
y0
 
∂
P (x) = Fy0 y0 = f (x) 0
∂y (1 + y 02 )1/2
(1 + y ) .1 − y 0 .y 0 /(1 + y 02 )1/2
02 1/2
 
= f (x)
1 + y 02
f (x)
= > 0 (since f (x) > 0) ,
(1 + y 02 )3/2
and
d
Q(x) = Fyy − Fyy0 = 0.
dx
Recall that the Jacobi Equation, with initial conditions, is
 
d du
P (x) − Q(x)u = 0 , u(a) = 0 , u0 (a) = 1 .
dx dx
The Jacobi equation is therefore
 
d du
P (x) = 0,
dx dx
so that
du c
= , where c is a constant.
dx P (x)
Now, at x = a, 1 = u0 (a) = c/P (a) so that c > 0 since P (a) > 0. Moreover, for
x > a, u0 (x) = c/P (x) > 0. Since u(a) = 0, it follows that u(x) > 0 for all
x > a. Hence there are no points conjugate to a on (a, b].
Since P (x) > 0, it follows that the stationary path is a weak local minimum of
the functional.
(c) Let y be the stationary path and let ỹ be any path satisfying the boundary
conditions ỹ(a) = A, ỹ(b) = B. Let h = ỹ − y. Then h(a) = h(b) = 0 and
ỹ = y + h. Then
Z b
S[ỹ] − S[y] = S[y + h] − S[y] = dx f (x)(1 + (y 0 + h0 )2 )1/2 −
a
Z b
dx f (x)(1 + y 02 )1/2
a
Z b n o
= dx f (x) (1 + (y 0 + h0 )2 )1/2 − (1 + y 02 )1/2
a
b
h0 y 0
Z
≥ dx f (x) (using the given inequality with z = y 0 ,
a (1 + y 02 )1/2
u = h0 and the property f (x) > 0)
Z b
= dx βh0 = β[h(b) − h(a)] = 0 (using ?? from part (a)
a
and h(a) = h(b) = 0.)
It follows that S[ỹ] ≥ S[y] and y gives a global minimum of S[y].

page 7 of ??