CHAPTER 1

Dimensional Analysis and Scaling

1. Dimensional analysis

Exercise 1.1 The variables are t, r, ρ, e, P . We already know one dimensionless

quantity π1 = ρr5 /et2 . Try to find another in terms of P , which is a pressure,
or force per unit area, that is, mass per length per time-squared. By inspection,
π2 = t2 P/r2 ρ is another dimensionless quantity. Thus we have
f (ρr5 /et2 , t2 P/r2 ρ) = 0
Now we cannot isolate the r and t variables in one dimensionless expression; they
must occur in both. If we solve for the first dimensionless quantity we get
ρr5 /et2 = F (t2 P/r2 ρ)
Then
1/5
et2

r= F (t2 P/r2 ρ)
ρ
Because the second dimensionless variable contains t and r in some unknown man-
ner, we cannot conclude that r varies like t2/5 . In this formulation, if one can argue
that the ambient pressure is small and can thus be neglected, then we can set P = 0
and obtain the result
 2 1/5
et
r= F (0)
ρ
which does imply that r varies like t2/5 .

Exercise 1.2 Here x = 0.5gt2 ; because g is an acceleration, we observe that π =

x/gt2 is dimensionless. Clearly this is the only dimensionless combination. So the
physical law is π = 0.5. The law could not depend on mass because there are no
dimensionless combinations of t, x, g that involve m.

2. The Buckingham Pi theorem

Exercise 2.1 Assume that f (v, Λ, g) = 0. If π is dimensionless

[π] = [v α1 Λα2 g α3 ]
= (LT −1 )α1 Lα2 (LT −2 )α3

1
2 1. DIMENSIONAL ANALYSIS AND SCALING

Thus we have the homogeneous system

α1 + α2 + α3 = 0, −α1 − 2α3 = 0
The rank of the coefficient matrix is one, so there is one dimensionless variable.
Notice that (−2, 1, 1) is a solution to the system, and thus
π = Λg/v 2
By the Pi theorem, F (π) = 0 or Λg/v 2 = Const.

Exercise 2.2 Follow the idea in Example 2.3 on page 11. Pick length, time and
mass as fundamental and write
x = λ1 x, t = λ2 t, m = λ3 m
Then write v = λ1 λ−1
2 v, and so on for the other variables. Show that
2 2 2
v − r2 ρ g µ−1 (1 − ρl /ρ) = λ1 λ−1
2 (v − r ρgµ
−1
(1 − ρl /ρ))
9 9
So, by definition, the law is unit free.

Exercise 2.6 We have A = f (c, φ) = c2 g(φ) since area is a length-squared. Now

apply the rule in the hint to the other two smaller, similar triangles to find their
areas as A1 = f (a, φ) = a2 g(φ) and A2 = f (b, φ) = b2 g(φ). Thus A = A1 + A2
implies c2 = a2 + b2 .

Exercise 2.7 The period cannot depend only on the length and mass. There
is no way that length and mass can be combined to yield a time dimension. If
we assume, however,
p that f (T, l, m, g) = 0, then there is only one dimensionless
variable,
p π = T g/l. (Again, mass will not appear). So, by the pi theorem,
T g/l =constant.

Exercise 2.9 Pick length L, time T , and mass M as fundamental dimensions.

Then the dimension matrix has rank three and there are√5 − 3 = 2 dimensionless
√
variables; they are given by π1 = γ and π2 = Rω ρl / P . Thus f (π1 , π2 ) = 0
implies
p
ω = R−1 P/ρl G(γ)
for some function G.

3. Scaling

Exercise 3.1 In (a) we have u = A sin ωt and so u′ = ωA cos ωt. Then M = A

and max |u′ | = ωA. Then we have tc = 1/ω. In (b) we have u = Ae−λt and u′ =
−λAe−λt . Then tc = max |u|/ max |u′ | = 1/λ. In part (c) we have u = Ate−λt and
u′ = (1−λt)Ae−λt . The maximum of u occurs at t = 1/λ and is M = A/λe. To find
the maximum of u′ we calculate the second derivative to get u′′ = Aλ(λt − 2)e−λt .
So the maximum derivative occurs at t = 2/λ or at an endpoint. It is easily checked
that the maximum derivative occurs at t = 0 and has value max |u′ | = A on the
given interval. Therefore tc = (A/λe)/A = 1/λe.
 3. SCALING 3

Exercise 3.2 Here u = 1 + exp(−t/ǫ) and u′ = −exp(−t/ǫ)/ǫ. Then tc =

max |u|/ max |u′ | = 2/ǫ−1 = 2ǫ. The time scale is very small, indicating rapid
change in a small interval. But a graph shows that that this rapid decrease occurs
only in a small interval near t = 0; in most of the interval the changes occur slowly.
Thus two time scales are suggested, one near the origin and one out in the interval
where t is order one.

Exercise 3.3 This exercise is a project on which students may work independently
or on a team. (a) The constant a must be budworms-squared since it is added to
such a term in the denominator. The entire predation term must be budworms
per time, and so b must have dimensions budworms per time. (b) The parameter a
defines the place where the predation term makes a significant rise. Thus it indicates
the threshold where the number of budworms is plentiful so that predation kicks
in; there are enough budworms to make the birds interested. (c) The dimensionless
equation is
dN N2
= sN (1 − N/q) −
dτ 1 + N2
(d) To find equilibrium solutions we set

N2
sN (1 − N/q) − =0
1 + N2
At this point we can use a calculator or a computer algebra program like Maple or
Mathematica to solve the equation for N . Observe that we obtain a fourth degree
polynomial equation when we simplify this algebraic equation:

sN (1 − N/q)(1 + N 2 ) − N 2 = 0

When s = 12 and q = 0.25 the equilibrium populations are N = 0, 0.261. When

s = 0.4 and q = 35 the equilibrium populations are N = 0, 0.489, 2.218, 32.29. The
plot in figure 1.1 shows the Maple generated solution curves in the case n(0) =
2, 25, 40.

−1
Exercise 3.4 There are p two possible time scales, one based on frequency ω , and
one based on period, l/g. If we scale by the inverse of frequency, we let τ = ωt.
Then the governing equation becomes
d2 θ
 
3g
+ − 1 sin θ = 0
dτ 2 2lω 2
p
If we scale by period, then pick the dimensionless time as T = t/ l/g. In this case
the governing equation becomes
d2 θ 3 lω 2
 
+ − sin θ = 0
dt2 2 g
p
When ω is small we should scale by l/g because that places the small term in the
equation, at the rotational force term, where it belongs; when ω is large we should
scale by ω −1 because then the small term appears in the gravitational force term,
where it belongs.
4 1. DIMENSIONAL ANALYSIS AND SCALING

Figure 1. Budworm populations for problem 3.3.

Exercise 3.5 To solve P ′ = P (1 − P ), P (0) = α we separate variables and

integrate using partial fractions:
1
Z Z
dP = dτ + c
P (1 − P )
or Z  
1 1
+ =τ +c
P 1−P
This gives
ln |P | − ln |p − 1| = τ + c
 3. SCALING 5

or
P
= Ceτ
P −1
The initial condition forces C = α/(α−1). This last expression can be manipulated
algebraically to obtain equation (23) in the text.

Exercise 3.6 Introduce the following dimensionless variables:

m = m/M, x = x/R, t = t/T, v = v/V
where T and V are to be determined. In dimensionless variables the equations now
take the form
αT
m′ = −
M
′ VT
x = − v
R
αβT 1 Tg
v′ = −
M V m V (1 − x)2
To ensure that the terms in the velocity and acceleration equations are the same
order, with the gravitational term small, pick
VT αβT
=
R MV
which gives
p
V = αβR/M
as the velocity scale.

Exercise 3.7 The differential equation is

q
c′ = − (ci − c) − kc2 , c(0) = c0
V
Here k is a volume per mass per time. Choosing dimensionless quantities via
C = c/ci , τ = t/(V /q)
the model equation becomes
dC
= −(1 − C) − bC 2 , C(0) = γ
dτ
where γ = c0 /ci and b = kV ci /q. Solve this initial value problem using separation
of variables or a computer algebra package.

Exercise 3.8 The quantity q is degrees per time, k is time−1 , and θ is degrees (one
can only exponentiate a pure number). Introducing dimensionless variables
T = T /Tf , τ = t/(Tf /q)
we obtain the dimensionless model
dT
= e−E/T − β(1 − T ), T (0) = α
dτ
where α = T0 /Tf , E = θ/Tf , and β = kTf /q.
6 1. DIMENSIONAL ANALYSIS AND SCALING

Exercise 3.9 Let h be the height measured above the ground. Then Newton’s
second law gives
mh′′ = −mg − a(h′ )2 , h(0) = 0, h′ (0) = V
Choose new dimensionless time and distance variables according to
τ = t/(V /g), y = h/(V 2 /g)
Then the dimensionless model is
y ′′ = −1 − α(y ′ )2 , y(0) = 0, y ′ (0) = 1
where prime is a τ derivative and α = aV 2 /mg.
p
Exercise 3.10 Scale time by l/g and angular displacement by θ0 . Introducing
the dimensionless variables
p
τ = t/ l/g, ψ = θ/θ0
gives the scaled model
d2 ψ
+ θ0−1 sin(θ0 ψ) = 0, ψ(0) = 1, ψ ′ (0) = 0
dτ 2
 CHAPTER 2

Perturbation Methods

1. Regular Perturbation

Exercise 1.1 Since the mass times the acceleration equals the force, we have
my ′′ = −ky − a(y ′ )2 . The initial conditions are y(0) = A, y ′ (0) = 0. Here, a is
assumed to be
p small. The scale for y is clearly the amplitude A. For the time
scale choose m/k,
p which is the time scale when no damping is present. Letting
y = y/A, τ = t/ m/k be new dimensionless variables, the model becomes
y ′′ + ǫ(y ′ )2 + y = 0
where ǫ ≡ aA/m. The initial data is y(0) = 1, y ′ (0) = 0. Observe that the small
parameter is on the resistive force term, which is correct.
Exercise 1.2 The problem is
u′′ − u = ǫtu, u(0) = 1, u′ (0) = −1
A two-term perturbation expansion is given by
1
y(t) = e−t + ǫ(et − e−t (1 + 2x + 2x2 ))
8
A six-term Taylor expansion is
1 1 − ǫ 3 1 − 2ǫ 4 1 − 4ǫ 5
y(t) = 1 − t + t2 + x + x − x
2 6 24 120
The plots in figure 1 show the superior performance of the two-term perturbation
approximation.
Exercise 1.3 (a) We have

t2 tanh t
= tanh t < 1
t2
for large t. Thus t2 tanh t = 0(t2 ) as t → ∞ .
(b) We have
e−t
lim =0
t→∞ 1
which proves the order relation.
(c) The order relation follows from the inequality
p
ǫ(1 − ǫ) √
√ = 1−ǫ≤1
ǫ
for small, positive ǫ.
7
8 2. PERTURBATION METHODS

Figure 1. Graphs of exact solution, Taylor series approximation,

and perturbation approximation in Exercise 1.2 when ǫ = 0.04.

(d) The idea is to expand cos ǫ in a power series to get

√
ǫ
1 − (1 − ǫ2 /2 + ǫ4 /4! − · · · )
√
ǫ
= 2
ǫ /2 − ǫ4 /4! + · · · )
1 1
= ǫ−3/2
2 1 − ǫ2 /12 + · · ·
But, using the geometric series,
1
= 1 + 0(ǫ2 )
1 − ǫ2 /12 + · · ·
which gives the result.
(e) We have
t 1
2
= ≤1
t t
for large t. So the ratio is bounded, proving the assertion.
(f) By Taylor’s expansion
eǫ − 1 = 1 + ǫ + 0(ǫ2 ) − 1 = 0(ǫ)
 1. REGULAR PERTURBATION 9

(g) Expand the integrand in a Taylor series and integrate term-by-term to get
Z ǫ Z ǫ
2 x2
e−x dx = (1 − x2 + + · · · )dx
0 0 2!
1
= ǫ − ǫ3 + · · ·
3
= 0(ǫ)
2 Rǫ 2
An alternate method is to notice that e−x ≤ 1, which gives 0 e−x dx ≤ ǫ. Thus
R ǫ −x2
0
e dx
≤1
ǫ

(h) Observe that

lim etan ǫ = 1
ǫ→0
Since the limit exists, the function must be bounded in a neighborhood of ǫ = 0,
which implies the result.
(i) Notice that
e−ǫ
= e−ǫ ǫp → 0
ǫ−p
as ǫ → ∞ (exponentials decay faster than power functions grow). One can use
L’Hospital’s rule to show this.
(j) Notice that
ln ǫ
= ǫp ln ǫ → 0
ǫ−p
as ǫ → 0 since power functions go to zero faster that logarithms grow near zero.
Again, use L’Hospital’s rule to verify this fact.

Exercise 1.4 Substitute x = x0 + ǫx1 + · · · into the nonlinear equation to get

h(ǫ) ≡ φ(x0 + ǫx1 + · · · , ǫ) = 0
By Taylor’s expansion
1
h(ǫ) = h(0) + h′ (0)ǫ + h′′ (0)ǫ2 + · · ·
2
Using the chain rule we can compute these derivatives of h at ǫ = 0 and thus expand
the equation in powers of ǫ. We get
φǫ (x0 , 0)
φ(x0 ) = 0, x1 = −
φx (x0 , 0)
and so on. To obtain x1 we clearly require φx (x0 , 0) 6= 0.

Exercise 1.5(a) Let τ = ωt, with ω = 1 + ω1 ǫ + · · · . Then the problem becomes

ω 2 y ′′ + y = ǫyω 2 (y ′ )2 , y(0) = 1, ωy ′ (0) = 0
Here, prime denotes a τ derivative. Now assume a regular perturbation expansion.
The leading order problem is
y0′′ + y0 = 0, y0 (0) = 1, y0′ (0) = 0
10 2. PERTURBATION METHODS

which has solution

y0 (τ ) = cos τ
The next order problem is
y1′′ + y1 = −2ω1 y0′′ + y0 (y0 )2 , y1 (0) = 0, y1′ (0) = 0
The equation simplifies to
1 1
y1′′ + y1 = ( + 2ω1 ) cos τ − cos 3τ
4 4
To eliminate the secular term take ω1 = −1/8. Then, to leading order,
  
1
y0 (t) = cos 1− ǫ t
8

Exercise 1.5(b) Let τ = ωt, with ω = 3 + ω1 ǫ + · · · . Then the problem becomes

ω 2 y ′′ + 9y = 3ǫy 3 , y(0) = 0, ωy ′ (0) = 1
Here, prime denotes a τ derivative. Now assume a regular perturbation expansion.
The leading order problem is
1
y0′′ + y0 = 0, y0 (0) = 0, y0′ (0) =
3
which has solution
1
y0 (τ ) = sin τ
3
The next order problem is
1
9y1′′ + 9y1 = 3y02 − 6ω1 y0′′ , y1 (0) = 0, y1′ (0) = −
9
The equation simplifies to
 
1 1 1
y1′′ + y1 = + 2ω1 sin τ − sin 3τ
9 12 9 · 36
To eliminate the secular term take ω1 = −1/24. Then, to leading order,
  
1 1
y0 (t) = sin 3− ǫ t
3 24

Exercise 1.6 The equation with ǫ can be handled with a perturbation series in ǫ.
After determining the coefficients, one can substitute ǫ = 0.001.

Exercise 1.7 If we ignor 0.01x in the first equation then y = 0.1. Then, from
the second equation x = 0.9. Checking these values by substituting back into the
first equation gives 0.01(0.9) + 0.1 = 0.109, so the approximation appears to be
good. But the exact solution is x = 190, y = 1. So the approximation is in fact
terrible. What went wrong? Since x = −90 the first term in the first equation
is 0.01x = −0.9, which is not small compared to the two other terms in the first
equation. Thus the first term cannot be neglected.
 1. REGULAR PERTURBATION 11

Exercise 1.10 The three-term approximation is given by

t4 t7 2
ya (t) = t + ǫ+ ǫ
12 504
The error is
t8
E(t, ǫ) = ya′′ − ǫtya = −ǫ3
504
Clearly the approximation is not uniform on t ≥ 0.

Exercise 1.11 The leading order solution is

√
y0 (t) = t(1 − ln t)
Substituting into the ODE gives
ǫ
Ly0 = − (1 + ln t)t3/2
2
We find
|Ly0 | ≤ 0.0448
Thus one would expect y0 to be a good approximation on 0 ≤ t ≤ e.

Exercise 1.12 Substitute the series u = u0 +u1 ǫ+· · · into the differential equation
and initial conditions:
1
u′0 + u′1 ǫ + · · · + u0 + u1 ǫ + · · · =
1 + u0 ǫ + · · ·
= 1 − u0 ǫ + 0(ǫ2 )
To get the last step we used the geometric series expansion. Now collecting the
coefficient of ǫ0 we get the leading order problem
u′0 + u0 = 1, u0 (0) = 0
Collecting the coefficients of ǫ we get
u′1 + u1 = −u0 , u1 (0) = 0
The solution to the leading order problem (a linear equation) is
u0 = 1 − e−t
The the next order problem becomes
u′1 + u1 = e−t − 1, u1 (0) = 0
This linear equation has solution
u1 = (t + 1)e−t − 1
Therefore, a two-term approximation is
u(t) = 1 − e−t + ǫ((t + 1)e−t − 1) + · · ·
12 2. PERTURBATION METHODS

2. Singular Perturbation

Exercise 2.1

(a) Consider the equation

ǫx4 + ǫx3 − x2 + 2x − 1 = 0
If x = O(1) then x2 − 2x + 1 ∼ 0 which means x ∼ 1, 1, a double root near
4 2
one. To find the remaining roots assume√ a dominant balance ǫx ∼ x with the
remaining terms small. Then x = O(1/ ǫ). This is a consistent balance because
ǫx4 , x2 = O(1/ǫ) and the remaining three terms are are order one and small in
comparison. So the dominant balance is
ǫx4 − x2 ∼ 0
√
which gives x ∼ ±1/ ǫ. So the leading order roots are
√
1, 1, ±1/ ǫ

(b) The equation

ǫx3 + x − 2 = 0
3
has an order one √ root x ∼ 2. The consistent3 dominant balance is ǫx ∼ x which
gives x = O(1/ ǫ). In this case we have ǫx + x ∼ 0, which gives the two other
leading order roots as
i
x ∼ ±√
ǫ
To find a higher order approximation for the root near x = 2 substitute x =
2 + x1 ǫ + · · · into the equation and collect coefficients of ǫ to get x1 = −8. Thus
x = 2 − 8ǫ + · · ·
√
Since the other two roots are are order O(1/ ǫ), let us choose a new order one
variable y given by
√
y = x/(1/ ǫ)
So, we are rescaling. Then the equation becomes
√
y3 + y − 2 ǫ = 0
and the√small term appears where it should in the equation. Now assume y =
y0 + y1 ǫ + · · · , substitute into the equation, and collect coefficients to get at
leading order
y03 + y0 = 0
√
which gives y0 = ±i. At order O( ǫ) we get the equation
3y02 y1 + y1 − 2 = 0
Thus y1 = −1 and we have the expansions
√
y = ±i − ǫ + · · ·
In terms of x,
i
x = ±√ − 1 + · · ·
ǫ
 2. SINGULAR PERTURBATION 13

(c) The equation

ǫ2 x6 − ǫx4 − x3 + 8 = 0
has three order one roots as solutions of −x3 + 8 = 0, or
√
x ∼ 2, −1 ± 3i
To find the other roots, the dominant balance is ǫ2 x6 − x3 ∼ 0, which gives
1 2πi/3 1 −2πi/3
x∼ e , 2/3 e
ǫ2/3 ǫ

(d) The equation

ǫx5 + x3 − 1 = 0
has three order one roots (the cube roots of one) given by
1 √
x ∼ 1, − ± 3i
2
The dominant balance for the remaining roots is between the first two terms which
gives
i
x ∼ ±√
ǫ

Exercise 2.2 The problem is

ǫy ′′ + y ′ + y = 0, y(0) = 0, y(1) = 0
When ǫ = 0 we get y ′ + y = 0 which has solution y = ce−t . This cannot satisfy
both boundary conditions so regular perturbation fails.
The characteristic equation is ǫm2 + m + 1 = 0 which has two real, negative
roots given by
1 √
m1 = (−1 + 1 − 4ǫ) = −1 + O(ǫ)
2ǫ
1 √ 1
m2 = (−1 − 1 − 4ǫ) = − + O(1)
2ǫ ǫ
√
Here we have used the binomial expansion 1 + x = 1 + x/2 + O(x2 ) for small x.
Note that one of the roots is order one, and one of the roots is large. So the general
solution is
y(t) = c1 em1 t + c2 em2 t
Applying the boundary conditions gives the exact solution
em1 t − em2 t
y(t) =
em1 − em2
Sketches of this solution show a boundary layer near t = 0 where there is a rapid
increase in y(t). Observe that em1 >> em2 . If t = O(1) then
em1 t ∼ e−t , em2 t ∼ 0
and thus
em1 t
y(t) ∼ ∼ e1−t
em1
14 2. PERTURBATION METHODS

which is an outer approximation. If t = O(ǫ), then

e(−1+O(ǫ))t − e(−1/ǫ+O(1))t
y(t) ∼
em1
eO(ǫ) − eO(ǫ) e−t/ǫ
∼
em1
−t/ǫ
1−e
∼
e−1
This is an inner approximation near t = 0.
Exercise 2.3 Observe that the equation can be written as a quadratic in ǫ:
2ǫ2 + xǫ + x3 = 0
Thus
1 p
ǫ= (−x ± x2 − 8x3 )
4
These two branches can be graphed on a calculator, and the graph shows that there
is just one negative value for x, near x = 0, in the case that ǫ is small and positive.
Thus assume
x = x1 ǫ + x2 ǫ2 + · · ·
Substituting into the given equation gives
x = −2ǫ + 8ǫ2 + · · ·

3. Boundary Layer Analysis

Exercise 3.2d The problem is

ǫy ′′ + (1 + t)y ′ + y = 0, y(0) = 0, y(1) = 1
By Theorem 3.1 in the text, there is a boundary layer at zero. The outer solution
is y0 = 2/(t + 1). In the boundary layer set τ = t/ǫ and Y (τ ) = y(t). Then the
inner equation is
Y ′′ + ǫτ + Y ′ + ǫY = 0
To leading order we have Yi′′ + Yi′ = 0 with Yi (0) = 0. So the inner approximation
is
Yi (τ ) = A(1 − e−τ )
Matching gives A = 2 and so the uniform approximation is
2
y(t) = + 2(1 − e−t/ǫ ) − 2
t+1

Exercise 3.2e The problem is

ǫy ′′ + t1/3 y ′ + y = 0, y(0) = 0, y(1) = e−3/2
There is a boundary layer near t = 0. The outer solution is
y0 (t) = exp(−1.5t2/3 )
In the inner region set τ = t/δ(ǫ). Then the dominant balance is between the first
and second terms and δ = ǫ3/4 . So the inner approximation to leading order is
Yi′′ + τ 1/3 Yi′ = 0
 3. BOUNDARY LAYER ANALYSIS 15

Solving gives
Z τ
Yi (τ ) = c exp(−0.75s4/3 )ds
0
√
Pick an intermediate variable to be η = t/ ǫ. Then matching gives
Z ∞ −1
c= exp(−0.75s4/3 )ds
0

Exercise 3.2g The problem is

ǫy ′′ + 2y ′ + ey = 0, y(0) = y(1) = 0
There is a layer at zero. The outer solution is
t+1
y0 (t) = − ln
2
In the boundary layer the first two terms dominate and δ(ǫ) = ǫ. The inner solution
is
Yi (τ ) = A(1 − e−2τ )
Matching gives A = ln 2. The uniform approximation is
t+1
y(t) = ln 2(1 − e−2t/ǫ ) − ln − ln 2
2

Exercise 3.2h The problem is

ǫy ′′ − (2 − t2 )y = −1, y(−1) = y(1) = 1
Now there are two layers near t = −1 and t = 1. The outer solution, which is valid
in the interval (−1, 1), away from the layers is
1
y0 (t) =
2 − t2
√
In the layer near t = 1 set τ = (1 − t)/δ(ǫ). We find δ = ǫ with inner equation,
to leading order, Yi′′ − Yi = 1. The inner solution is
Yi (τ ) = 1 + ae−τ − (1 + a)eτ
√
In the layer near t = −1 set τ = (t − 1)/δ(ǫ). We find δ = ǫ with inner equation,
to leading order, (Yi∗ )′′ − Yi∗ = −1. The inner solution is
Yi∗ (τ ) = 1 + be−τ − (1 + b)eτ
Matching gives a = b = 1 and the uniform approximation is
1 √ √
y(t) = 2
− e(t−1)/ ǫ − e(t+1)/ ǫ
2−t

Exercise 3.5 The problem is

1
ǫy ′′ + y ′ + y = 0, y(0 + 1, y ′ (0) = 0
t
which is an initial value problem and appears to be singular. But, the outer solution
is
2
y0 (t) = Ce−t /2
16 2. PERTURBATION METHODS

and it is observed that it satisfies both initial conditions when C = 1. It also

satisfies the ODE uniformly, i.e.,
1 2
ǫy0′′ + y0′ + y0 = ǫ(t2 − 1)e−t /2 = O(ǫ)
t
Thus it provides a uniform approximation and the problem does not have a layer.
It is instructive to try to put a layer at t = 0; one finds that no scaling is possible.
Exercise 3.7a By Theorem 3.1 the problem
ǫy ′′ + (1 + t2 )y ′ − t3 y = 0, y(0) = y(1) = 1
has a layer of order δ = ǫ at t = 0. The outer solution is
r
2 t2 − 1
y0 (t) = 2
exp( )
1+t 2
The inner solution is
Yi (τ ) = A + (1 − A)e−τ
p
Matching gives A = 2/e.
Exercise 3.7b By Theorem 3.1 the problem
ǫy ′′ + (cosh t)y − −y = 0, y(0) = y(1) = 1
has a layer of width δ = ǫ near t = 0. The outer solution is
y0 (t) = exp(2 arctan et − 2 arctan e)
In the layer use the expansion
1
cosh z = 1 + z 2 + · · ·
2
The inner approximation is
Yi (τ ) = 1 − A + Ae−τ
Matching gives A = 1 − exp(π/2 − 2 arctan e).
Exercise 3.7c The problem
2ǫ
ǫy ′′ + y ′ − y = 0, y(0) = 0, y ′ (1) = 1
t
If we assume a layer at t = 1 the outer solution is y0 (t) = 0. In the inner region
near t = 1 the dominant
√ balance is between the first and last terms and the width
of the layer is δ = ǫ. The inner variable is τ = (1 − t)/δ. The inner solution is
√
Yi (τ ) = ( ǫ + b)e−τ + beτ
We must set b = 0 to stay bounded. So the uniform solution is
√ √
y(t) = ǫe(t−1)/ ǫ

Exercise 3.7d In this problem we have an outer solution y0 (t) = 0 which satisfies
both boundary conditions exactly. So we have an exact solution y ≡ 0.
Exercise 3.7e The problem is
1 ′
ǫy ′′ + y + ǫy = 0, y(0) = 0, y(1) = 1
t+1
 4. TWO APPLICATIONS 17

The outer approximation is y0 (t) =const, so there are several possibilities. The
idea that works is to assume a layer at t = 0 of width δ = ǫ which gives the inner
approximation
Yi′′ (τ ) = a(1 − e−τ )
Matching gives a = 1.

4. Two Applications

Exercise 4.1 The equation my ′′ + ay ′ + by = 0 has characteristic roots

a 1 p 2
λ± = − ± a − 4mk
2m 2m
Because m is small, the roots are real. The general solution is
y(t) = c1 eλ− t + c2 eλ+ t
Applying the initial conditions y(0) = 0, my ′ (0) = I, we get the exact solution
I
eλ+ t − eλt


y(t) = √
2
m a − 4mk
Now, if m is small, then λ− ≈ −a/m and λ+ ≈ 0. Thus, for small m, the exact
solution is approximately
I
y(t) ≈ (1 − e−at/m )
a
which agrees with the approximate solution in equation (14) in the text.

Exercise 4.2 If we make the change of variables

p
t = t/ m/k, y = y/(I/a)
then we obtain the dimensionless initial value problem
ǫy ′′ + y ′ + ǫy = 0, y(0) = 0, y ′ (0) = ǫ−1
where √
ǫ = km/a << 1
Then, to leading order (setting ǫ = 0 in the differential equation) we have y ′ (t) = 0,
or y(t) = const. = 0. So the outer approximation is zero, which is valid after a long
time.

Exercise 4.3 (a) In this case the system is

x′ = y − ǫ sin x, ǫy ′ = x2 y + ǫy 3
with initial conditions x(0) = k, y(0) = 0. Setting ǫ = 0 we get the outer equations
x′0 = y0 , 0 = x20 y0
Here we can choose y0 = 0 and x0 = k and the initial conditions are met. So this
problem does not have an initial layer. It is a regular perturbation problem with
leading order solution
x0 (t) = k, y0 (t) = 0
It is instructive for the student to assume a layer near t = 0 and carry out the
analysis to find that the inner approximation agrees with the outer approximation.
18 2. PERTURBATION METHODS

(b) The problem is

u′ = v, ǫv ′ = u2 − v, u(0) = 1, v(0) = 0
Assume a boundary layer near t = 0. Then the outer problem is
u′0 = v0 , v0 = −u20
Then
u′0 = −u20
which has solution (separate variables)
1 −1
u0 (t) = , v0 (t) =
t+c (t + c)2
In the boundary layer take η = t/ǫ. Then the inner problem is
U ′ = ǫV, V ′ = −U 2 − V, U (0) = 1, V (0) = 0
Thus, setting ǫ = 0 and solving yields the inner approximation
U (η) = const. = 1, V (η) = e−η − 1
Matching gives
lim u0 (t) = lim U (η)
t→0 η→∞

or 1/c = 1. Hence, c = 1. Then the uniform approximation is

1 1
u= +1−1=
t+1 t+1
and
−1 −1
v= 2
+ e−t/ǫ − 1 − (−1) = + e−t/ǫ
(t + 1) (t + 1)2

5. The WKB Approximation

√
Exercise 5.1 Letting ǫ = 1/ λ we have
ǫ2 y ′′ − (1 + x2 )2 y = 0, y(0) = 0, y ′ (0) = 1
This is the nonoscillatory case. From equation (10) on p 88 of the text the WKB
approximation is, after applying the condition y(0) = 0
√ Z x √ Z x
    
c1 2 2 2 2
yW KB = exp λ (1 + ξ ) dξ − exp − λ (1 + ξ ) dξ
1 + x2 0 0
√ Z x
 
2c1
= sinh λ (1 + ξ 2 )2 dξ
1 + x2 0

Applying the condition y ′ (0) = 1 gives c1 = 1/2.

√
Exercise 5.2 Letting ǫ = 1/ λ we have
y ′′ + λ(x + π)4 y = 0, y(0) = y(π) = 0
This is the oscillatory case and Example 5.3 on page 89 applies. The large eigen-
values are !
nπ 9n2
λn = R π =
0
(x + π)2 dx 49π 4
 5. THE WKB APPROXIMATION 19

The eigenfunctions are

 Z x 
C 3n 2
sin (ξ + π) dξ
x+π 7π 2 0
3n x3
 
C 2 2
= sin ( + πx + π x)
x+π 7π 2 x

Exercise 5.3 The solution is a straightforward substitution.

Exercise 5.4 This problem is solved exactly like Exercise 5.1.
Exercise 5.5 Make the change of variables t = ǫx. Then the differential equation
becomes
d2 y
ǫ2 2 + q(t)2 y = 0
dt
which is equation (11) on page 88. So the approximation is given by equation (12)
on page 89 with x replaced by ǫx.
We can think of the equation y ′′ + q(ǫx)2 y = 0 as a harmonic oscillator where
the frequency is q(ǫx), which is time-dependent. (Here, think of x as time). If ǫ is
small, it will take a large time x before there is significant changes in the frequency.
Thinking of it differently, if q(x) is a given frequency, then the graph of q(ǫx) will
be stretched out; so the frequency will vary slowly.
√
Exercise 5.6 Here we apply formula (12) on page 89 with k(x) = e2x and ǫ = 1/ λ.
Then the WKB approximation is
√ ! √ !
c1 λ 2x c2 λ 2x
yW KB = x sin (e − 1) + x cos (e − 1)
e 2 e 2
Applying the condition y(0) = 0 gives c2 = 0. Then
√ !
c1 λ 2x
yW KB = x sin (e − 1)
e 2
Then y(1) = 0 gives
√ !
λ
sin (e − 1) = 0
2
and this forces
4π 2
λ=
(e − 1)2
for large n.
Exercise 5.9
(a) We have mx′′ = F (x). We use prime to denote time derivatives. Mul-
tiplying by x′ and noting that x′ x′′ = 12 ((x′ )2 )′ gives m ′ 2 ′ ′
2 ((x ) ) = F (x)x . But
m ′ 2 ′
V (x) = (dV /dx)x = −F (x)x , and so it follows that 2 ((x ) ) − V (x)′ = 0.
′ ′ ′

Integrating then gives

m ′ 2
(x ) − V (x) = E
2
where E is a constant of integration.
20 2. PERTURBATION METHODS

(b) Letting p = mx′ in the last equation and then solving for p gives
p
p = ± 2m(E − V (x))

6. Asymptotic Expansion of integrals

Exercise 6.2 Making the substitution t = tan2 θ gives

Z π/2
1 ∞ e−λt dt
Z
−λ tan2 θ
I(λ) = e dθ = √
0 2 0 (1 + t) t
Now Watson’s lemma (Theorem 6.1) applies. But we proceed directly by expanding
1/(1 + t) in its Taylor series
1
= 1 − t + t2 − · · ·
1+t
which gives
1 ∞ e−λt
Z
I(λ) = √ (1 − t + t2 − · · · )dt
2 0 t
Now let u = λt. This gives
Z ∞ √
u u3/2
 
1 −u 1
I(λ) = √ e √ − + 2 + ···
2 λ 0 u λ λ
Then, using the defintion of the gamma function,
1 1 1 3 1 5
I(λ) = √ (Γ( ) − Γ( ) + 2 Γ( ) + · · · )
2 λ 2 λ 2 λ 2

Exercise 6.3 Assume that g has a maximum at b with g ′ (b) > 0. Then expand
g(t) = g(b) + g ′ (b)(t − b) + · · ·
The integral becomes
Z b
I(λ) = f (t)eλg(t) dt
a
Z b
′
= f (t)eλ(g(b)+g (b)(t−b)+··· ) dt
a
Z b
′
λg(b)
≈ f (b)e eλg (b)(t−b) dt
a
Now make the substitution v = λg ′ (b)(t − b) to obtain
Z 0
1
I(λ) ≈ f (b)eλg(b) ′ ev dv
λg (b) λg′ (b)(a−b)
or
f (b)eλg(b) 0 v
Z
I(λ) ≈ e dv
λg ′ (b) −∞
or
f (b)eλg(b)
I(λ) ≈
λg ′ (b)
for large λ.
 6. ASYMPTOTIC EXPANSION OF INTEGRALS 21

If the maximum of g occurs at t = a with g ′ (a) < 0, then it is the same

calculation. We expand g in its Taylor series about t = a and we obtain the same
solution except for a minus sign and the b in the last formula replaced by a.

Exercise 6.4 (a) We have, using a Taylor expansion,

Z ∞
I(λ) = e−λt ln(1 + t2 )dt
0
Z ∞
t4 t6
= e−λt (t2 − + − · · · )dt
0 2 3
Now let u = λt and we get
1 ∞ −u u2 u4 u6
Z
I(λ) = e ( 2 − 4 + 6 + · · · )dt
λ 0 λ 2λ 3λ
Using the definition of the gamma function, we obtain
1 2! 4! 6!
I(λ) = ( 2 − 4 + 6 + · · · )
λ λ 2λ 3λ

(b) Let g(t) = 2t − t2 . This ′

√function has its maximum at t = 1 where g (1) = 0
′′
and g (1) = −2. Take f (t) = 1 + t. Then
Z 1 s
√ λ(2t−t2 ) 1 λg(1) −2π
I(λ) = 1 + te dt ≈ f (1)e ′′ (1)
0 2 λg
r
π λ
= e
2λ

(c) Let g(t) = 1/(1 + t). This function has its maximum,
√ with a negative
derivative, at t = 1. Thus Exercise 6.3 holds. With f (t) = 3 + t we have
f (1)eλg(1) 8
I(λ) ≈ − = eλ/2
λg ′ (1) λ

Exercise 6.5 We have Z ∞

Γ(x + 1) = ux e−u du
0
Integrate by parts by letting r = ux and ds = e−u du. Then the integral becomes
Z ∞ Z ∞
x −u
u e du = x e−u ux−1 du = xΓ(x)
0 0

Exercise 6.6 (b) Using the fact that e−t < 1 for t > 0, we have
Z ∞ −t
e
|rn )λ)| = n! n+1
dt
λ t
Z ∞
1
≤ n! dt
λ tn+1
−1 1
= n! n |∞ = (n − 1)! n → 0
nt λ λ
22 2. PERTURBATION METHODS

as λ → ∞.

(c) Observe that

∞
e−t e−λ
Z
dt ≤
λ tn+1 λn+1
Then the ratio of rn to the last term of the expansion is
|rn (λ)| n
−λ n
≤
(n − 1)!e /λ λ
This tends to zero as λ → ∞. So the remainder is little oh of the last term, and so
we have an asymptotic series.

(d) Fix λ. The nth term of the series is does not converge to zero as n → ∞.
Therefore the series does not converge.

Exercise 6.7 We have ∞

1
Z
I(λ) = e−t dt
0 t + λ)2
We integrate by parts letting
1
u= , dv = e−t
t + λ)2
We get Z ∞
1 1
I(λ) = 2 − 2 e−t dt
λ 0 t + λ)3
Now integrate by parts again via
1
u= , dv = e−t
t + λ)3
Then Z ∞
1 2 1
I(λ) = 2 − 3 + 6 e−t dt
λ λ 0 t + λ)4
Continuing in the same manner gives
1 2! 3!
I(λ) = − 3+ 2
λ2 λ λ Z ∞
n! 1
+ · · · + n−1 (−1)n+1 + (n + 1)! e−t dt
λ 0 t + λ)n+2
 CHAPTER 3

Calculus of Variations

1. Variational Problems

Exercise 1.1 The functional is

Z 1
J(y) = (y ′ sin πy − (y + t)2 )dt
0

First note that if y(t) = −t then J(y) = 2/π. Now we have to show that J(y) < 2/π
for any other y. To this end,
Z 1
J(y) = (y ′ sin πy − (y + t)2 )dt
0
Z 1
≤ y ′ sin πydt
0
1 1
Z
= − (cos πy)′ dt
π 0
1 2
= − (cos πy(1) − cos πy(0)) ≤
π π

Exercise 1.2 Hint: substitute

y(x) = x + c1 x(1 − x) + c2 x2 (1 − x)
into the functional J(y) to obtain a function F = F (c1 , c2 ) of the two variables
c1 and c2 . Then apply ordinary calculus techniques to F to find the values that
minimize F , and hence J. That is, set ∇F = 0 and solve for c1 and c2 .

2. Necessary Conditions for Extrema

Exercise 2.1 (a) The set of polynomials of degree ≤ 2 is a linear space. (b) The
set of continuous functions on [0, 1] satisfying f (0) = 0 is a linear space. (c) The set
of continuous functions on [0, 1] satisfying f (0) = 1 is not a linear space because,
for example, the sum of two such functions is not in the set.

Exercise 2.2 We prove that

Z b
||y||1 = |y(x)|dx
a
23
24 3. CALCULUS OF VARIATIONS

is a norm on the set of continuous functions on the interval [a, b]. First
Z b Z b
||αy||1 = |αy(x)|dx = |α| |y(x)|dx = |α| ||y||1
a a
Rb
Next, if ||y||1 = 0 iff a
|y(x)|dx = 0 iff y(x) = 0. Finally, the triangle inequality if
proved by
Z b Z b
||y + v||1 = |y(x) + v(x)|dx ≤ (|y(x)| + |v(x)|)dx
a a
Z b Z b
= |y(x)|dx + |v(x)|dx = ||y||1 + ||v||1
a a
The proof that the maximum norm is, in fact, a norm, follows in the same
manner. To prove the triangle inequality use the fact that the maximum of a sum
is less than or equal to the sum of the maximums.

Exercise 2.3 Let y1 = 0 and y2 = 0.01 sin 1000x. Then

||y1 − y2 ||s = max |0.01 sin 1000x| = 0.01
and
||y1 − y2 ||w = max |0.01 sin 1000x| + max |(0.01)(1000) cos 1000x| = 10.01

Exercise 2.4 We have

J(y0 + ǫαh) − J(w)
δJ(y0 , αh) = lim
ǫ→0 ǫ
J(y0 + ǫαh) − J(w)
= lim α
ǫ→0 ǫα
J(y0 + ηh) − J(w)
= lim α
η→0 η
= αδJ(y0 , h)

Exercise 2.5 (a) not linear; (b) not linear; (c) not linear; (d) not linear; (e) linear;
(f) not linear.

Exercise 2.6 An alternate characterization of continuity of a functional that is

often easier to work with is: a functional J on a normed linear space with norm
|| · || is continuous at y if for any sequence of functions yn with ||yn − y|| → 0 we
have J(yn ) → J(y) as n → 0.
(a) Now let ||yn − y||w → 0. Then ||yn − y||s → 0 (because ||v||s ≤ ||v||w ). By
assumption J is continuous at y in the strong norm, and therefore J(yn ) → J(y).
So J is continuous in the weak norm.
Rbp
(b) Consider the arclength functional J(y) = a 1 + (y ′ )2 dx. If two functions
are close in the strong norm, i.e., if the maximum of their difference is small, then
it is not necessarily true that their arclengths are close. For example, one may
oscillate rapidly while the other does not. Take y = 0 and y = n−1 sin nx for large
n. These two functions are close in the strong norm, but not the weak norm.
 2. NECESSARY CONDITIONS FOR EXTREMA 25

Exercise 2.7 Let yn → y in the strong norm. Then

Z b
|J(yn ) − J(y)| ≤ ||yn + y||s ||yn − y||s dx = (b − a)||yn + y||s ||yn − y||s
a
But ||yn + y||s ≤ M for some constant M since the sequence converges. Thus
|J(yn ) − J(y)| → 0.

Exercise 2.8 R
b
(a) δJ(y, h) = a (hy ′ + yh′ )dx.
Rb
(b) δJ(y, h) = a (2h′ y ′ + 2h)dx.
(c) δJ(y, h) = ey(a) h(a).
(d) See Exercise 3.6.
Rb
(e) δJ(y, h) = a h(x) sin x dx.
Rb
(f) δJ(y, h) = a 2y ′ h′ dx + G′ (y(b))h(b).

Exercise 2.9 Let yn → y. Then J(yn ) = J(yn − y) + J(y) → J(y) because

J(yn − y) → 0 (since yn − y → 0 and J is continuous at zero by assumption).

Exercise 2.10 We have

Z b
x(y ′ + ǫh′ )2 + (y + ǫh) sin(y ′ + ǫh′ )2 dx


J(y + ǫh) =
a
We must take the second derivative of this function of ǫ with respect to ǫ and then
set ǫ = 0. We obtain
Z b
δ2 J = (2x(h′ )2 − y(h′ )2 sin y ′ + 2hh′ cos y ′ )dx
a

Exercise 2.11 Here the functional is

Z 1
J(y) = (x2 − y 2 + (y ′ )2 )dx
0
Then Z 1
δJ(y, h) = (−2yh + 2y ′ h′ )dx
0
Thus
3
δJ(x, x2 ) =
2
and
∆J = J(y + ǫh) − J(y) = J(x + ǫx2 ) − J(x) = etc.

Exercise 2.13 In this case

Z 2π
J(y) = (y ′ )2 dx
0
Then Z 2π Z 2π
J(y + ǫh) = (1 + ǫ cos x)2 dx = 2π + ǫ2 cos2 xdx
0 0
26 3. CALCULUS OF VARIATIONS

Then
2π
d
Z
J(y + ǫh) = 2ǫ cos2 xdx
dǫ 0
and so
d
J(y + ǫh) |ǫ=0 = 0
dǫ
Thus, by definition, J is stationary at y = x in the direction h = sin x. The family
of curves y + ǫh is shown in the figure.

Exercise 2.14 Here Z 1

J(y) = (3y 2 + x)dx + y(0)2
0
Then Z 1
δJ(y, h) = 6yh dx + 2y(0)h(0)
0
Substituting y = x and h = x + 1 gives δJ = 5.

3. The Simplest Problem

Exercise 3.1 (a) The Euler equation reduces to an identity (0 = 0), and hence
every C 2 function that satisfies the boundary conditions is an extremal. (b) The
Euler equation reduces to an identity (0 = 0), and hence every C 2 function that
satisfies the boundary conditions is an extremal. (c) The Euler equation reduces to
y = 0, which does not satisfy the boundary conditions; so there are no extremals.

Exercise 3.2 (a) The Euler equation is

d d
Ly − Ly′ = (2y ′ /x3 ) = 0
dx dx
Thus
y(x) = Ax4 + B
(b) The Euler equation is
y ′′ − y = ex
The general solution is
x x
y(x) = Aex + Be−x + e
2

Exercise 3.3 The Euler equation is

d p d f y′
Ly − Ly′ = fy 1 + (y ′ )2 − p =0
dx dx 1 + (y ′ )2
p
Taking the total derivative and then multiplying by 1 + (y ′ )2 gives
f (y ′ )2 y ′′
fy − y ′ fx − f y ′′ + =0
1 + (y ′ )2
This reduces to
f y ′′
fy − y ′ fx − =0
1 + (y ′ )2
 3. THE SIMPLEST PROBLEM 27

Exercise 3.4 The Euler equation is

d
− Ly′ = 0
dx
or
y′
p =c
x 1 + (y ′ )2
Solving for y ′ (take the positive square root since, by the boundary conditons, we
want y ′ > 0), separating variables, and integrating yields
Z r 2 2
c x
y(x) = dx + k
1 − c2 x2
Then make the substitution u = 1 − c2 x2 to perform the integration. We obtain
1p
y(x) = 1 − c2 x2 + k
c
Applying the boundary conditions to determine the constants finally leads to
p
y(x) = − 5 − x2 + 2
which is an arc of a circle.

Exercise 3.6 Let h ∈ C 2 . Then

Z bZ b
δJ(y, h) = K(s, t)[y(s)h(t) + h(s)y(t)]dsdt
a a
Z b Z b
+2 y(t)h(t)dt − 2 h(t)f (t)dt
a a
Now, using the symmetry of K and interchanging the order of integration allows
us to rewrite the first integral as
Z bZ b
2 K(s, t)y(s)h(t)dsdt
a a
Then !
Z b Z b
δJ(y, h) = 2 K(s, t)y(s)ds + y(t) − f (t) h(t)dt
a a

Thus
Z b
K(s, t)y(s)ds + y(t) − f (t) = 0
a
This is a Fredholm integral equation (see Chapter 5) for y.

Exercise 3.7 The Euler equation is

−((1 + x)y ′ )′ = 0
or
y ′ = c1 /(1 + x)
Integrating again
y(x) = c1 ln(1 + x) + c2
28 3. CALCULUS OF VARIATIONS

If y(0) = 0, y(1) = 1 then we get

ln(1 + x)
y(x) =
ln 2
If the boundary condition at x = 1 is changed to y ′ (1) = 0, then the extremal is
y(x) ≡ 0.

Exercise 3.8 In each case we minimize the functional T (y) given in equation (11)
in the text. For example, in (a) we have n = ky and so the Euler equation reduces
to
ky
p =c
1 + (y ′ )2
Solving for y ′ and then separating variables gives
dy
±dx = p 2
C1 y 2 − 1
This integrates to
1
±x + c2 = cosh−1 (c1 y)
c1
So y is a hyperbolic cosine.

Exercise 3.9 Observe that

d dL dLy′
Lt − (L − y ′ Ly′ ) = Lt − + y′ + y ′′ Ly′
dt dt dt
dLy′
= Lt − Lt − Ly y ′ − Ly′ y ′′ + y ′ + y ′′ Ly′
dt
d
= −y ′ (Ly − Ly′ )
dt

Exercise 3.10 The minimal surface of revolution is found by minimizing

Z b p
J(y) = 2πy 1 + (y ′ )2 dx
a
Because the integrand does not depend explicitly on x, a first integral is
L − y ′ Ly′ = c
Upon expanding and simplifying, this equation leads to
dy p
= k2 y2 − 1
dx
Now separate variables and integrate while using the fact that
d 1
cosh−1 u = √
du 2
u −1

Exercise 3.11 The Euler equation becomes

x2 y ′′ + 2xy ′ − y = 0
which is a Cauchy-Euler equation. Its solution is
√ √
y(x) = Ax(−1+ 5)/2
+ Bx(−1− 5)/2
 4. GENERALIZATIONS 29

(recall that a Cauchy-Euler equation can be solved by trying power functions, y =

xm for some m).

4. Generalizations

Exercise 4.1 By way of contradiction, assume f is strictly positive at some point

x = c. By continuity, f is positive in some open interval I = (c − ǫ, c + ǫ). Take
h(x) = (x − c + ǫ)5 (c + ǫ − x)5 in I and zero otherwise. Having the fifth power
on these factors makes h four times continuously differentiable. This gives the
Rb
contradiction since a h(x)f (x)dx > 0.

Exercise 4.2 (a) The Euler equations are

8y1 − y2′′ = 0, 2y2 − y1′′ = 0
Eliminating y2 gives
y1′′′′ − 16y1 = 0
The characteristic equation is m4 −16 = 0, which has roots m = ±2, ±2i. Therefore
the solution is
y1 (x) = ae2x + be−2x + c cos 2x + d sin 2x
where a, b, c, d are arbitrary constants. Then y2 = 0.5y1′′ , and the four constants
a, b, c, d can be computed from the boundary conditions.
(b) The Euler equation is
y ′′′′ = 0
and thus the extremals are
y(x) = a + bx + cx2 + dx3
The four constants a, b, c, d can be computed from the boundary conditions.

L(x, y, y ′ , y ′′ )dx is
R
Exercise 4.3 The Euler equation for j(y) =
Ly − (Ly′ )′ + (Ly′′ )′′ = 0
If Ly = 0 then clearly
Ly′ − (Ly′′ )′ = const.
If Lx = 0, then expand all the derivatives and use the Euler equation to show
d
(L − y ′ (Ly′ − (Ly′′ ) − y ′′ Ly′′ )) = 0
dx

Exercise 4.4 (a) The Euler equation simplifies to y ′′′′ = 0 and thus the extremals
are
y(x) = a + bx + cx2 + dx3
The four constants a, b, c, d can be computed from the boundary conditions.
(b) The Euler equation simplifies to
y ′′′′ + y = 0
30 3. CALCULUS OF VARIATIONS

The characteristic equation is m4 + 1 = 0, which has the four complex roots

1 1 1 1
m = √ (1 + i), √ (1 − i), √ (−1 + i), √ (−1 − i)
2 2 2 2
Therefore the general solution is
√ √
   
x/ 2 x x −x/ 2 x x
y(x) = e a cos √ + b sin √ +e c cos √ + d sin √
2 2 2 2
Observe that a = b = 0 in order to meet the boundary conditions at x = +∞. The
boundary conditions at x = 0 determine c and d.

Exercise 4.5 (a) The Euler equation is

(x2 ux )x + (y 2 uy )y = 0
(b) The Euler equation is
utt − c2 uxx = 0
which is the wave equation.

Exercise 4.8 The Euler equation is

utt − m2 u = ∆u
where ∆u = uxx + uyy + uzz is the Laplacian operator.

Exercise 4.9 The Euler equation is

y ′′′′ − 2y ′′ + y = 0
This constant coefficient equation has characteristic equation m4 −2m2 +1 = 0 with
roots m = ±1, ±1. So, both 1 and −1 are double roots. Therefore the extremals
are
y(x) = aex + bxex + ce−x + dxe−x

Exercise 4.10 The two Euler equations (expanded out) are

Ly′ y′ y ′′ + Ly′ z′ z ′′ = 0
Lz′ y′ y ′′ + Ly′ z′ z ′′ = 0
It is given that the determinant of the coefficient matrix of this system is nonzero.
Therefore the only solution is y ′′ = z ′′ = 0, which gives linear functions for y and
z.

Exercise 4.12 (a) The Euler equation is

y ′′ − y = 0
which gives
y(x) = aex + be−x
One boundary condition is y(0) = 1. The natural boundary condition is y ′ (1) = 0.
These two conditions determine a and b.
 4. GENERALIZATIONS 31

(b) The Euler equation reduces to y ′′ − 1 = 0 which leads to

1
y(x) = x2 + ax + b
2
The natural boundary condition is y ′ (1) + y(1) + 1 = 0. This condition, with the
given condition y(0) = 0.5 determines a and b.

Exercise 4.13 The Euler equation is

1
x2 y ′′ + 2xy ′ + y = 0
4
which is a Cauchy-Euler equation. Assuming solutions of the form y = tm gives
the characteristic equation
1
m(m − 1) + 2m + = 0
4
which has a real double root m = − 21 . Thus the general solution is
1 1
y(x) = a √ + b √ ln x
x x
The given boundary condition is y(1) = 1; the natural boundary condition at x = e
is y ′ (e) = 0. One finds from these two conditions that a = b = 1.

Exercise 4.14 The first variation of the functional is

Z Z
δJ(u, h) = E(x, y)dxdy + (−hLuy dx + hLux dy)
R C
where E is the Euler expression. Pick h such that h = 0 on C. Then E = 0 by the
fundamental lemma. Thus
Z
(−hLuy dx + hLux dy) = 0
C
for all h. But this last condition forces
−Luy dx + Lux dy = 0
on C or
(Lux , Luy ) · n = 0
where n is the outward unit normal to C.

Exercise 4.15 Using Exercise 4.14 we find that the natural boundary condition is
a∇u · n = 0 on the curve Bd R.

Exercise 4.16 The natural boundary condition is

Ly′ (b, y(b), y ′ (b)) + G′ (y(b)) = 0

Exercise 4.17 Using the last problem the natural boundary condition is y ′ (1) +
y(1) = 0. The extremal is
1
y(x) = − x + 1
2
.
32 3. CALCULUS OF VARIATIONS

5. Hamiltonian Theory

Exercise 5.1 The Hamiltonian is

p2
H(t, y, p) = − q(t)y 2
4r(t)
Hamilton’s equations are
p
y′ = , p′ = 2q(t)y
2r(t)

Exercise 5.2 Here we have

Z p
J(y) = (t2 + y 2 )(1 + (y ′ )2 )

We find p
t2 + y 2 y ′
p = Ly′ = p
1 + (y ′ )2
which yields
p2
(y ′ )2 =
t 2 + y 2 − p2
The Hamiltonian simplifies, after some algebra, to
p
H(t, y, p) = − t2 + y 2 − p2
Then Hamilton’s equations are
dy p dp y
=p , =p
dt t + y 2 − p2
2 dt t + y 2 − p2
2

Dividing the two equations gives

dy p
=
dp y
Integrating yields
y 2 − p2 = const
These are hyperbola in the yp phase plane.

Exercise 5.3 Hamilton’s equations for the pendulum are

p
θ′ = , p′ = −mgl sin θ
ml2

Exercise 5.4 (a) The potential energy is the negative integral of the force, or
1 1
Z
V (y) = − (−ω 2 y + ay 2 )dy = ω 2 y 2 − ay 3
2 3
The Lagrangian is L = 21 m(y ′ )2 −V (y). The Euler equation concides with Newton’s
second law:
my ′′ = −ω 2 y + ay 2
(b) The momentum is p = Ly′ = my ′ , which gives y ′ = p/m. The Hamiltonian
is
1
H(y, p) = (p/m)2 + V (y)
2
 5. HAMILTONIAN THEORY 33

which is the kinetic plus the potential energy, or the total energy of the system.
Yes, energy is conserved (L is independent of time).
(c) We have H = E for all time t, so at t = 0 we have
1 ω2
(p(0)/m)2 + V (y(0)) =
2 10
If y(0) = 0 we can solve for p(0) to get the momentum at time zero; but this gives
p
y ′ (0) = ± ω 2 /5m
(d) and (e) The potential energy has a local maximum at y = ω 2 /a and is equal
to
ω6
Vmax = V (ω 2 /a) =
6a2
2
Note that V = 0 at y = 0, 3ω /2a. If E < Vmax then we obtain oscillatory motion;
if E > Vmax , then the motion is not oscillatory. Observe that the phase diagram
(the solution curves in yp–space, or phase space), can be found by graphing
p
p = ± 2m(E − V (y)
for various constants E.

Exercise 5.6R Here the force is F (t, y) = ket /y 2 . We can define a potential by
t
V (t, y) = − F (t, y)dy = ke /y. The Lagrangian is
m
L(t, y, y ′ ) = (y ′ )2 − V (t, y)
2
The Euler equation is
my ′′ − ket /y 2 = 0
which is Newton’s second law of motion. The Hamiltonian is
p2 k
H(t, y, p) = + et
2m y
which is the total energyy. Is energy conserved? We can compute dH/dt to find
dH
= ket /y 6= 0
dt
So the energy is not constant.

Exercise 5.7 The kinetic energy is T = m(y ′ )2 /2. The force is mg (with a plus
sign since positive distance is measured downward). Thus V (y) = −mgy. Then
the Lagrangian is L = m(y ′ )2 /2 + mgy.

Exercise 5.9 The Lagrangian does not depend explicitly on the independent vari-
able x and therefore a first integral can be calculated from
L − y ′ Ly′ = C

Exercise 5.10 Pick r and θ as generalized coordinates. Then T = (m/2)((r′ )2 +

r2 (θ′ )2 ). There are two contributions to the potential energy, one due to gravity
and the other due to the spring. We have
k
V (y) = mg(l − r cos θ) + (l − r)2
2
34 3. CALCULUS OF VARIATIONS

Therefore the Lagrangian is

m ′ 2 k
((r ) + r2 (θ′ )2 ) − mg(l − r cos θ) − (l − r)2
L(r, θ, r′ , θ′ ) =
2 2
The two equations of motion are
(r2 θ′ )′ + gr sin θ = 0, mr′′ − mr(θ′ )2 − mg cos θ − k(l − r) = 0

Exercise 5.11 We have r′ = −α and so r(t) = −αt + l. Then the kinetic energy is
m m
T = ((r′ )2 + r2 (θ′ )2 ) = (α2 + (l − αt)2 (θ′ )2 )
2 2
and the potential energy is
V = mgh = mg(l − r cos θ) = mg(l − (l − αt) cos θ)
The Lagrangian is L = T − V . The Euler equation, or the equation of motion, is
g sin θ + (l − αt)θ′′ − 2αθ = 0
One can check that the Hamiltonian is not the same as the total energy; energy is
not conserved in this system. If fact, one can verify that
d
(T + V ) = −mgα cos θ 6= 0
dt

Exercise 5.14 The Emden-Fowler equation is

2
y ′′ + y ′ + y 5 = 0
t
2
Multiply by the integrating factor t to write the equation in the form
(t2 y ′ )′ + t2 y 5 = 0
Now we can identify this with the Euler equation:
Ly = −t2 y 5 , Ly′ = t2 y ′
Integrate these two equations to find
1 1
L = t2 (y ′ )2 − t2 y 6 + φ(t)
2 6

Exercise 5.15 Multiply y ′′ + ay ′ + b = 0 by the integrating factor exp(at) to get

(eat y ′ )′ + beat = 0
Now identify the terms in this equation with the terms in the Euler equation as in
Exercise 5.14. Finally we arrive at a Lagrangian
m 
L = eat (y ′ )2 − by
2
There are many Lagrangians, and we have chosen just one by selecting the arbitrary
functions.

Exercise 5.16 Follow example 5.6 in the book with m, a, k and y replaced by
L, R, C −1 and I, respectively.
 6. ISOPERIMETRIC PROBLEMS 35

Exercise 5.17 For which Lagrangians is the Euler equation satisfied identically for
every y? The Euler equation is
Ly − Lty′ − Lyy′ y ′ − Ly′ y′ y ′′ = 0
IF this equation holds for every function y then we can pick y so that the first three
terms vanish, and thus
Ly′ y′ = 0
Hence Ly′ = φ(t, y), where φ is an arbitrary function. Hence,
L = φ(t, y)y ′ + ψ(t, y)
where ψ is another arbitrary function.

6. Isoperimetric Problems

Exercise 6.1 Let L∗ = x2 + (y ′ )2 + λy 2 . Then the Euler equation for L∗ is

y ′′ + λy = 0, y(0) = y(1) = 0
If λ ≥ 0 then this boundary value problem has only trivial solutions. If λ < 0, say
λ = −β 2 , then the problem has nontrivial solutions
yn (x) = Bn sin nπx, n = 1, 2, . . .
where the Bn are constants. Applying the constraint gives
Z 1
Bn2 sin2 nπx dx = 2
0
Thus Bn = ±2 for all n. So the extremals are
yn (x) = ±2 sin nπx, n = 1, 2, . . .

Exercise 6.2 The Euler equations become

d ∗ d ∗
L∗y1 − L = 0, L∗y2 − L =0
dx y1 dx y2
where L∗ = L + λG.

Exercise 6.3 Form the Lagrangian

p
L∗ = xy ′ − yx′ + λ (x′ )2 + (y ′ )2
Since the Lagrangian does not depend explicitly on t, a first integral is given by
L∗ − x′ L∗x′ − y ′ L∗y′ = c

Exercise 6.4 The problem is to minimize

Z 1p
J(y) = 1 + (y ′ )2 dx, y(0) = y(1) = 0
0
subject to the constraint
Z 1
y(x)dx = A
0
36 3. CALCULUS OF VARIATIONS

Form the Lagrangian p

L∗ = 1 + (y ′ )2 + λy
Since the Lagrangian does not depend explicitly on x, a first integral is given by
L∗ − y ′ L∗y′ = c
Expanding out this equation leads to
s
1 − (λy − c)2
y′ =
(λy − c)2
Separating variables and integrating, and then using the substitution u = 1 − (λy −
c)2 , gives
1 du
Z
− √ = x + c1
2λ u
Thus
1
(x + c1 )2 + (y − c/λ)2 = 2
λ
which is a circle. Now the two boundary conditions and the constraint give three
equations for the constants c, c1 , λ.

Exercise 6.5 Form the Lagrangian

L∗ = p(y ′ )2 + qy 2 + λry 2
Then the Euler equation corresponding to L∗ is
Ly − (Ly′ )′ = 2qy + 2rλy − 2(py ′ )′ = 0
or
(py ′ )′ − qy = rλy, y(a) = y(b) = 0
This is a Sturm-Liouville problem for y (see Chapter 4).

Exercise 6.6 Solve the constraint equation to obtain

z = g(t, y)
Substitute this into the functional to obtain
Z b Z b
W (y) ≡ F (t, y, y ′ ) ≡ L(t, y, g(t, y), y ′ , gt + gy y ′ )dt
a a
Now form the Euler equation for F . We get
d
Fy − Fy ′ = 0
dt
or, in terms of L,
d
Ly + Lz gy + Lz′ (gty + gyy y ′ ) − (Ly′ + Lz′ gy ) = 0
dt
This simplifies to
d d
Ly − Ly′ + gy (Lz − Lz′ = 0
dt dt
Also G(t, y, g(t, y)) = 0, and taking the partial with respect to y gives
Gy + Gz gy = 0
 6. ISOPERIMETRIC PROBLEMS 37

Thus
d d
Ly − dt Ly′ Lz − dt Lz′
=
Gy Gy
Now these two expressions must be equal to the same function of t, that is
d d
Ly − Ly′ = λ(t)Gy , Lz − Lz′ = λ(t)Gz
dt dt

Exercise 6.7 Form the Lagrangian

L∗ = (y ′ )2 + λy 2
The Euler equation reduces to
y ′′ − λy = 0, y(0) = y(π) = 0
This is similar to Exercise 6.1. The extremals are given by
p
yn (x) = ± 2/π sin nx, n = 1, 2, 3, . . .
38 3. CALCULUS OF VARIATIONS

Figure 1. Exercise 1

Additional Exercises

1. (See the figure below) A drop of a viscous fluid with viscosity µ, measured in
units of grams/(sec cm), and density ρ is sitting on a flat plate, making contact
area A with the plate. The drop has volume V . The fluid drop is surrounded by
an inviscid (viscosity zero) fluid of the same density ρ which is moving with speed
U . Let τ be the time taken for the drop to move. (a) Use dimensional analysis to
determine a functional relation for τ in terms of a minimal number of parameter
combinations. (b) Suppose it is known that τ is inversely proportional to µ. Does
this additional information allow a more precise characterization of the functional
form for τ ?

2. Consider the following differential equation for u = u(t), t > 0, where a is a

positive real number:
du
= (1 − u)(u2 − a)
dt
(a) find all the equilibrium solutions and (b) determine a condition on a that will
guarantee that u = 1 is an attractor.

3. Consider the initial value problem

d2 y
+ 9y = 3ǫy 3 , y(0) = 0, y ′ (0) = 1
dt2
where 0 < ǫ << 1. Find the leading order perturbation approximation using the
Poincare-Lindstedt method (make sure your approximation contains the adjusted
frequency).

4. Prove that
3x2 ǫ2
= O(ǫ2 )
x2 + 1
as ǫ → 0, uniformly on x ≥ 0.
 6. ISOPERIMETRIC PROBLEMS 39

5. Find a two-term perturbation approximation on 0 ≤ x ≤ 1 for the solution y(x)

of
1
y′ + y = √ , y(0) = 0, 0 < ǫ << 1
1+ǫ y
.

6. Write down an order relation with big oh or little oh that expresses the fact that
e−x decays faster than 1/x2 as x → +∞, and then prove your assertion.

4
7. Consider the equation
√ 0.0001x − 0.01x + 1 = 0. To leading order, determine
1
the roots. (Recall that i = ± √2 (1 + i)).

8. A functional J, defined on the set A of all y ∈ C 2 [0, 1] with y(0) = y(1) = 0, is

defined by
Z 1p
J(y) = 1 + (y ′ )2 dx
0
Let h ∈ C 2 [0, 1] with h(0) = h(1) = 0. Compute δJ(y, h).

9. Find a one-term approximation for the large eigenvalues λ of the problem

y ′′ + λ2 e2x y = 0, 0 < x < 1, y(0) = y(1) = 1

10. Find a leading order approximation on 0 ≤ x ≤ 1 to the solution of the problem

1
ǫy ′′ + 2y ′ + y 3 = 0, y(0) = 0, y(1) =
2

11. Consider the initial value problem

ǫe−t u′′ + u′ = 1 − 2t, t > 0, u(0) = 1, ǫu′ (0) = −1
where 0 < ǫ ≪ 1 and u = u(t). Find a leading order approximation for the solution.

12. Find the extremal for the functional

Z 1
J(y) = (1 + x)(y ′ )2 dx, y(0) = 0, y(1) = 1
0

13. Find a two-term approximation for all the real roots of the equation eǫx = x2 −1
where 0 < ǫ ≪ 1.
R1 3
14. Find the leading order behavior of the integral 0
e−λt dt for large λ.

15. Consider the following functional defined over the set of C 2 functions satisfying
y(0) = 1 and y(b) = 0:
Z b
J(y) = (G(y)(y ′ )2 − G(y)2 )dx
0
where G is a given differentiable function.
40 3. CALCULUS OF VARIATIONS

Figure 2. Exercise 16

(a) Write out the Euler equation in its most simplified form (take the total
derivative and simplify completely).
(b) Find an expression ψ(y, y ′ ) that is constant whenever y is a solution to the
Euler equation, and subsequently show that the extremal must be given implicitly
by the formula
Z ys
G(η)
x=± dη
1 c − G(η)2

for some constant c to be determined.

16. A particle of mass m is attached to two springs as it moves along the y-axis (see
the figure). Let y = y(t) be its displacement from equilibrium, where y = 0, with
positive displacement measured to the right. The force due to the spring on the
left is linear and proportional to the displacement, and the force due to the spring
on the right is nonlinear and proportional to the cube root of the displacement.
Initially the mass is displaced a units to the right of equilibrium and then released.
(a) Using Newton’s second law, formulate an initial value problem that governs
the displacement of the mass.
(b) Carefully formulate Hamilton’s principle for this system.
(c) In the case that the force due to the linear spring is much smaller than the
force due to the nonlinear spring, the problem can be scaled and reduced to the
dimensionless problem

u′′ + ǫu + u1/3 = 0, u(0) = 1, u′ (0) = 0

where ǫ is a small positive parameter. Assume a regular perturbation expansion

and write down (but do not solve) the O(1) and O(ǫ) problems.

17. Find a one-term perturbation approximation to the boundary value problem

ǫu′′ + (x + 1)u′ + u = 2x, 0 < x < 1, 0 < ǫ ≪ 1

u(0) = 1, u(1) = 2
 6. ISOPERIMETRIC PROBLEMS 41

18. Consider the functional defined by

Z b
J(u) = p(x)2 (u′ )2 dx, u(a) = A, u(b) = B; p ∈ C[a, b], p>0
a
Find an explicit formula for the extremal.

19. Find a three-term asymptotic approximation to the function

Z x
2
I(x) = e−s dx
0
as x → 0.

20. Consider the problem

ǫu′′ + b(t)u′ + u = 0, 0 < t < T ; 0 < ǫ ≪ 1
′
u(0) = 0, u (0) = β/ǫ + γ
where b is smooth and b(t) > 0, and T, β, γ > 0. Find a uniformly valid, leading
order approximation to this problem.