3

Non-linear ordinary differential

equations

3.1 Introduction
The differential equations met in Chapter 2 were linear in the sense
that they were special cases of the general nth order linear equation

where fly f2, . . . , fn, g are given functions of x. For such equations the
Principle of Superposition applies: an arbitrary linear combination of
individual solutions is also a solution.
Equations which cannot be written in the form of (3.1) are called
non-linear and, for such equations, the Principle of Superposition does
not apply. A typical first-order non-linear equation is

whereas
d2y , dy

is a second-order non-linear equation.

There is no general method of solving non-linear equations analyti-
cally and numerical procedures are frequently the only techniques
available. However, some limited types can be solved analytically by
special methods. These are discussed in subsequent sections.
53
54 Non-linear ordinary differential equations

3.2 Equations with separable variables

In general, the first-order equation
dy/dx=f(x,y)/g(x,y), (3.4)
where f(x, y) and g(x, y) are given functions, will be non-linear.
However, if f(x, y) and g(x, y) are separable so that f(x, y) =
X(x)Y(y) and g(x, y) = U(x)V(y), then
dy/dx = F(x)/G(y), (3.5)
where F(x) = X(x)/U(x) and G(y) = V(y)/Y(y). If we multiply (3.5)
by G(y) and integrate both sides with respect to x, we obtain

\
G(y) dy
"I dx + C (3.6)

where C is a constant.

Example 1
dy CO

dx
(3.7)
X

Separating the variables gives

I sec2 ydy=i — + C, (3.8)

where C is a constant. Carrying out the integrations and writing

C = \nA, where A is another constant, we obtain
tany =\nx + \nA (3.9)
so that
y = tan-1[ln(^jc)]. ^ (3.10)

A case of a first-order non-linear equation which can be reduced to

separable form is
dy/dx = K(y/x), (3.11)
where K is a given function of y/x only. Writing y = xu(x) gives

x^ + u = K(u), (3.12)
ax
or
^ =^ ^ . (3.13)
dx x
This equation is now of separable form.
 3.2 Equations with separable variables 55

Example 2
d JL
i
dx x—y'
(3-14)
Writing (3.14) in the form of (3.11), we have

dy _ y/x
(3.15)
dx 1 —y/x
Putting y = xu gives
du u
(3.16)
~ dx 1-u'
or
du u2
(3.17)
dx x{\ — u)

Separating the variables leads to

f — ^ d u = I — + ln,4, (3.18)
J u J x
where A is a constant. Integrating both sides, we have

- - - I n n = ln(i4*), (3.19)

so that the relationship between x and y is, on substituting for u=y/x,

(y) ( ) (y) O. ^ (3.20)

We note that in Example 2 we cannot determine y explicitly in terms

of x. Although the separation of variables and integration can always,
in principle, be carried out for equations of the type (3.5), there is no
guarantee that y may be found explicitly in terms of x.
It may be possible, after a first integration, to reduce a second-order
non-linear equation to a first-order separable type. We now give an
example to illustrate this.

Example 3

xg-(,-0* (3.21,
given that y = 1 and dy/dx = 1 at x = 0.
56 Non-linear ordinary differential equations

Since

'(&.& +* (3.22)
dx \ dxl dx dx
then (3.21) can be written

x w
dx\ dx) dx " 'dx' "~'
or, simplifying,

d I dy\ dy d , ~
dx \ dx) dx dx

Integrating with respect to x gives

2
X^=$y + A, (3.25)

where A is a constant. Applying the boundary conditions, we find that

A = — \y so that (3.25) becomes

x-^ = \(y2-\). (3.26)

dx
Separating the variables gives

^ = j j + \nB, (3.27)

where B is a constant. The left-hand side can be integrated by

expressing (y2 — I)" 1 in partial fractions. We have

(3 28)
ii^iJTi) ' -
or

Hence
y - l = Bx(y + l), (3.30)
or
 3.3 Equations reducible to linear form 57

To find B, we again apply the boundary conditions. The condition

y = 1 at x = 0 is satisfied by (3.31), and to satisfy the condition
dy/dx = 1 at x = 0 we need to find dy/dx. From (3.31),
2B
£ (3 32)
dx (\-Bxf
At x = 0, 1 = 2B, giving B = 3. The complete solution, from (3.31), is
therefore

3.3 Equations reducible to linear form

In some cases a non-linear equation may be reduced to linear form
and hence solved by standard techniques. The substitution p = dy/dx
may help to carry out this reduction.

Example 4 Consider

Differentiating with respect to x, we have

dy d2y\ I dy\ dy d2y dy
dx / \ dx! dx dx dx
Eliminating the logarithm between (3.34) and (3.35) gives, on multi-
plying through by y + dy Idx,
dyd2y (dy\2 dy
-T^ri + A^) +y-r = ®- ( 3 - 36 )
2
dxdx \dx) dx
Hence either dy/dx = 0 or d2y/dx2 + 2 dy/dx + y = 0, both of which
are linear equations. The possible solutions are therefore
y=C (3.37)
and
y = e~x(Ax + B), (3.38)
where A, B and C are constants.
We now determine the values of these constants so that (3.37) and
(3.38) satisfy (3.34). Inserting y = C into (3.34) gives
ClnC + C = 0, (3.39)
58 Non-linear ordinary differential equations

so that C = 0 or C = e~1. In the case C = 0 the expression C l n C is

undefined. Hence the solution is C = e" 1 giving

y = e" 1 . (3.40)
For y = e~x(Ax + B), y + dy/dx = Ae~x. Inserting these into (3.34),
we have
*) + e~x(Ax + B) = 0. (3.41)
Cancelling e~* and using \n(Ac~x) = In A -x, we find
A(\n A - x) + Ax + B = 0, (3.42)
so that
,41n,4 + £ = 0. (3.43)
Substituting back for B> we have a solution

y = e~x(Ax - A In A), (3.44)

where A is an arbitrary constant which can be found by imposing a
boundary condition on y. ^4

Example 5 Consider

where y = \ when x = 0 and dy/dx = \ when y — 0. Putting

p= dy/dx, (3.46)
we have
d2y/dx2=pdp/dy. (3.47)
Now (3.45) becomes
pdp/dy + 2p2 = y2. (3.48)
This can be written

^ ( P 2 ) + 2p2 = / > (3.49)

which is a linear first-order equation in p2. Putting z=p2 in (3.49),

we find
4z = 2y2, (3.50)
which may be solved by multiplying by the integrating factor e4y.
 3.4 Bernoulli's equation 59

Accordingly,

— (ze4y) = 2y2e4y, (3.51)

dy
and hence

ze4y = 2\ y2Q4y dy + C, (3.52)

where C is a constant. Performing the integration and dividing by

e4y gives
(3.53)

From the boundary conditions, z = (dy/dx)2 = je when y = 0. Hence

C = 0 and
P 2 = b> 2 -b> + * . (3.54)
Taking the square root,

P = f = ^V[(r-l)2 + ^ (3.55)
the positive sign being chosen so that p = \ when y = 0. The variables
may now be separated giving

(3.56)

where A is a constant. Hence

V2 sinh'M y-T1) =x + A (3.57)

Now y = i at x = 0, so A = 0. The solution for y is therefore

y = \[l + sinm>/V2)]. ^ (3.58)

3.4 Bernoulli's equation

A linear equation of the type

(3.59)
may be solved, as in Example 5 above, by multiplying by the
integrating factor exp(J P(x) dx). The non-linear equation

dy/dx+ P(x)y=y"Q(x), (3.60)

60 Non-linear ordinary differential equations

where ni=0, 1, is often called Bernoulli's equation. Using the

substitution
Hyn~l = z, (3.61)
then

and (3.60) becomes

~ + P(x)z = Q(x). (3.63)

1 — ndx
This is now of the linear form (3.59) and hence may be solved by the
integrating factor method.

Example 6 Consider
dy/dx + 2y=xy3. (3.64)
Hence

1 ^ + 1 = *, (3.65)
y dx y
and therefore
1d/1 \ 1.
2 (3.66)
2dx\y ) y2
Writing z = 1/y2, we have
dz/dx-4z = -2x. (3.67)
Multiplying by the integrating factor e~4x, we find

e~4xz = -2 [xe~Axdx + Cy (3.68)

where C is a constant. Performing the integration and multiplying by

e4* gives

z = \ = & + i + Ce4x. ^ (3.69)

Sometimes it is not obvious how an equation can be converted into

Bernoulli form. It may be helpful to invert the equation, that is, to
regard it as an equation for x in terms of y (where y is the independent
variable and x is the dependent variable). To illustrate this, we give
the following example.
 3.5 Riccati's equation 61

Example 7 Consider

x(l - xy2) — + 2y = 0, (3.70)

whence

dx x{\-xy2Y {iJl)

This would seem to be difficult to solve but, by inverting, we obtain

dx _ x(l— xy2) _ x x2y

dy 2y 2y 2
and so

=y (3 73)
j+Y ~T- *
We see that (3.73) is an equation of Bernoulli type when viewed as an
equation for x in terms of y. Letting z = 1/x, then

= (3 74)
T~T ~o> '
dy 2y 2
for which the integrating factor is y~K Hence

zy~± = -[ -y-t dy + C, (3.75)

giving
(3.76)

The relationship between x and y is therefore

-=-&2 + Cyi. ^ (3.77)

x

3.5 Riccati's equation

Riccati's equation is

dy/dx =p(x)y2 + q(x)y + r(x), (3.78)

where /?, ^ and r are given functions of x. Suppose one solution of
(3.78) is known, say y = S(x). We then put

y = S(x) + l/u(x). (3.79)

62 Non-linear ordinary differential equations

From (3.79) and the differential equation (3.78), we obtain

*- + ^ . (3.80)
dx u dx
But, since S is a solution of (3.78),
dS/dx =p(x)S2 + q(x)S + r(x), (3.81)
and (3.80) becomes
du/dx + [2p(x)S + q(x)]u + p(x) = 0. (3.82)
This equation is of standard first-order linear form which may be
solved by the integrating factor method. Here the integrating factor is

v = exp{ J [2p(x)S(x) + q(x)] dx}, (3.83)

and consequently u(x) may be obtained.

Example 8 Consider

Here y = 1 = S(x) is a solution. Hence putting

y = 1 + 1/M, (3.85)
as in (3.79), we find
du/dx + (2 + l/x)u = - 1 . (3.86)
For this equation the integrating factor is

exp[[ (2 + IIx) dx] = xe2*, (3.87)

whence

= j-x^dx +Q (3.88)
where C is a constant. Integrating and dividing by xe 2

(3.89)

From (3.85) finally

1
(3.90)
 3.6 Special forms of RiccatVs equation 63

3.6 Special forms of Riccati's equation

1.
In (3.78) we consider first the case of p(x) = — 1 so that
dyldx +y2 = q(x)y + r(x). (3.91)
Writing

so that

T- = - T 4 - A ( ? ) > (3-93)
dx zdx z \dxl
(3.91) becomes 2
dz dz
—2q{x)
-q(x)--r(x)z=0. (3.94)
d?- d-x
This is a linear second-order equation for z and appropriate methods
may now be applied (for example, the Frobenius series method used in
Chapter 2 if q and r are simple polynomials). We note that if
p(x) = 4-1 in (3.78), then the substitution

will give the equation

d2z / dz ,x

—2 - q{x) — 4- r{x)z = 0. (3.96)

This is also a linear equation.

2.
Secondly we consider the case of q(x) = 0 in (3.78) so that
dyldx =p(x)y2 4- r(x). (3.97)
We now make a change of the independent variable x to x', where xr
is defined by
dxf/dx=p(x)y (3.98)
or

x'= \p(x)dx. (3.99)

Then
dy dy dx' dy
dx dx' dx dx'
64 Non-linear ordinary differential equations

Hence from (3.97) we have

p{x) — = p{x)y2 + r(x). (3.101)

Equation (3.101) may be written as

dy/dxr=y2+f(x'), (3.102)
where
f{x') = r[x{x')]lp[x{x')} (3.103)
is found using the relationship (3.99) between x and x' (this may be
difficult to obtain in many cases since x' is explicitly given in terms of
x, but not vice-versa). Equation (3.102) may be transformed to linear
form since it is a Riccati equation with p(x) = +l as discussed in
1 above.

Example 9 Consider
dy/dx = 2xy2 - 2x3. (3.104)
This is a case of (3.78) with q{x) = 0. We therefore change the variable
from x to

x' = J2xdx=x2, (3.105)

following (3.99). Then

% = %dfx=2x% (3 106)
-
and substituting this into (3.104) and cancelling 2x gives
dy/dx' = y2 - x2 = y2 - x'. (3.107)
f
This is now a Riccati equation for y in terms of x' with p(x ) in (3.78)
equal to +1. Following (3.95), we put
1 dz
y—- (3.108)

and obtain from (3.107)

—4 = *'*- (3.109)
This is Airy's equation discussed in Section 2.9 and hence a solution
for z in terms of x' may be found in terms of Airy functions. The
solution for y is then obtained from (3.108) and finally substituting x2
for*' from (3.105). ^
 3.7 The Lane-Emden equation 65

3.7 The Lane-Emden equation

The form of this equation is

d2y 2 dy
dx xdx

where a is a constant. Solutions which satisfy the boundary conditions

y = l, dy/dx = 0 at x = 0, (3.111)

are called Lane-Emden functions. Exact solutions are known when

a = 0, 1 and 5. In the cases a = 0 and a = 1, the equation is linear and
the solutions are easily found. When a = 5, let

Then after some algebra we find

4d2u/dt2-u + u5 = 0. (3.113)

Now writing v = du/dt, so that d2u/dt2 = v dv/du, (3.113) becomes

4vdv/du - M + M5 = 0. (3.114)

This equation may be integrated with respect to u to give

2v2-\u2 + Ub=C, (3.115)

where C is a constant. From the boundary conditions (3.111) and the

transformations (3.112), we see that x = 0 corresponds to f—»oo and
hence u and du/dt must tend to zero as t—»o°. Hence u = v=0 as
/—>oo and so, from (3.115), C = 0 giving
U2 = iw2_x^ (3.116)
or
v = du/dt = £MV(l - 5"4)- (3.117)
Separating the variables and integrating, we have

where A is a constant. To evaluate the integral on the left-hand side,

we put u4 = 3 cos2 6 so that

t + A = - J sec 6 d& = -ln(sec 6 + tan 6). (3.119)

66 Non-linear ordinary differential equations

Taking the exponential of both sides yields, with B = e~A,

Be1 = sec 0 + tan 0. (3.120)
Using cos 0 = w2/V3 and x = e~', we find
(3 m)
$V(^) -
the solution of which is

From (3.112), the solution for y is

V$B
l
Applying the boundary condition (3.111) that y = 1 when x = 0 gives
5 = 1/V3 and, finally,
y = 1/V(1 + k 2 )- (3.124)

3.8 The non-linear pendulum

The non-linear equation
d2d/dt2 + o)2 sin 0 = 0, (3.125)
where a> is a constant, is called the non-linear pendulum equation and
may be solved in terms of a class of integrals known as elliptic
integrals. Multiplying (3.125) by 2dd/dt and integrating, we have
{dOldtf = 2co2 cos 0 + C, (3.126)
where C is a constant of integration. Assume, for example, that
dd/dt = 0 when 0 = a. Then C = —2(o2 cos a and
(dd/dtf = 2co2(cos 0 - cos a). (3.127)
Taking the square root, separating the variables and integrating, we
have
d6
w\l2 Jo V(cos 0 - cos a)'
assuming 0 = 0 when t = 0. If t = T at 0 = 6lt then

a>V2J<> V(cos 0 - cos a)'

 3.8 The non-linear pendulum 67

Using
cos 0 = 1 - 2 sin2 (0/2), cos a = 1 - 2 sin2 (or/2), (3.130)
(3.129) becomes

2a> Jo V[sin 2 (ar/2) - sin

sir2(<9/2)]' v

Now writing
sin(0/2) = £ s i n 0 , (3.132)
where k = sin(ar/2), then

in which k sin 0X = sin(0!/2). The integral in (3.133) is referred to as

an elliptic integral of the first kind and is usually denoted by

(3 134)
'w4>)' "
with O^A;^1. Similarly, the elliptic integral of the second kind is
defined by

E(k, 0 0 = f ' V(l - k2 sin2 0) d(j>y (3.135)

where, as before, 0 ^ k ^ 1. In both cases the elliptic integrals are said

to be complete if 0 t = JT/2 and are then denoted by F(k) and E(k)>
respectively. Graphs of F{k, 0 0 and E(k, 0 0 are shown in Figures
3.1(a) and 3.1(b). We see that (3.133) gives T explicitly in terms of 0X
but not vice-versa.
We mention here that other integrals may be expressed in terms of
the above elliptic integrals and hence evaluated by consulting tables of
values (see reference on page 22).

Example 10 Evaluate

d<t>
1
Jo V(l - 4 sin
Putting 4 sin2 0 = sin2 6, we obtain
rjz/2
68 Non-linear ordinary differential equations

Example 11 Evaluate
dx
(3.138)

Putting x = sin 0, we obtain

rjr/6

h J rn

0
cos
(3.139)

However, 3 — 4 sin 2 <p + sin 4 <p = (3 — sin 2 (p){\ — sin 2 <p) = (3 — sin 2
cos 2 <p. H e n c e

_d$ = _L-/ 1 (3.140)

h = -rrz
- 3 sin 0) v3

3.9 Buffing's equation

In its simplest form Duffing's equation is
d2y/dx2 + ay + by3 = 0, (3.141)
where a and b are constants. We proceed, as in the case of the

Figure 3.1

k=0

(a)
 3.9 Buffing's equation 69

non-linear pendulum equation, by multiplying (3.141) by 2dy/dx.

Then

This equation may be integrated directly to give

=C-ay2-lby\ (3.143)

where C is a constant of integration. Taking the square root of (3.143)

and separating the variables, we find

wrx+B- <3144)
where B is another constant. For particular values of a, b and C, the
left-hand side can be expressed in terms of an elliptic integral.

Example 12 If
f l = - l y + y \ (3.145)
with boundary conditions y = 0 and dy/dx = l when x = 0, find the
x-value for which y = 1.
Proceeding as above, we have

= 2(—\y +y ) — , (3.146)

which on integration gives

= C-ly2 + hy\ (3.147)
where C is a constant. Since y = 0 when dy/dx = 1, we find C = 1 and
^AkWa-iZ + l/). (3.148)
Separating the variables and using y = 0 when JC = 0,
dy
f (3.149)

The value of x, JC0 say, for which y = 1 is

3?2. i..4,- (3-150)

Now putting y = sin <py we have dy = cos 0 d0 and

1 " b2 + 2 / = 2 ( 2 - 3 sin2 0 + sin4 0) = (1 - \ sin2 (p) cos2 0. (3.151)
70 Non-linear ordinary differential equations

Hence

— J sin <p)
using the definition of the complete elliptic integral of the first
kind. ^

In general, it is extremely difficult to obtain analytic (closed form)

solutions to Duffing's equation. However, a technique known as the
phase-plane method is useful in finding the nature of the solutions of
(3.141) for any a and b values and given initial conditions. We define
dy/dx = w, (3.153)
so that (3.141) becomes
dw/dx = -ay-by3. (3.154)
The equations (3.153) and (3.154) are now a particular case of the
general system
dy/dx = P(y,w),}
(3.155)

where, for Duffing's equation,

P(y, w) = w, Q(y, w) = -ay - by3. (3.156)
It is for a system of the form (3.155) that the phase-plane method is
appropriate. By eliminating x from (3.155), we have
P(y, w)
(3.157)
dw Q{y,w)'
In particular, for the Duffing equation,
dy w
(3.158)
dw ay + by3'
Separating the variables and integrating, we find the solution of
(3.158) to be
\ay2 + \ C, (3.159)
where C is a constant determined by the initial conditions. The
(y, w) = (yy dyldx) plane is called the phase-plane and the nature of
the solution (3.159) of (3.158), as represented by the curves of y and w
in the phase-plane for different values of C, yields some information
about the solution of Duffing's equation, but does not give the solution
 3.9 Buffing's equation 71

itself. For example, it can be shown that the existence of simple

(non-intersecting) closed curves in the phase-plane implies the exist-
ence of periodic solutions of the original equation. Specifically,
Duffing's equation will have periodic solutions if there exist simple
closed (y, w) curves described by (3.159). To illustrate this, we
consider the following values of a and b:

1. a = l, b=2
In this case (3.159) becomes
/ + y2 + w2 = 2C. (3.160)
The graph of this family of curves is shown in Figure 3.2. From
(3.160), we see that it is not possible to have C < 0 for any initial
conditions. For any value of C ^ 0, the curve is closed, which indicates
that only periodic solutions of the equation exist for these values of a
and b.

2. a = 1, b = - 2
In this case (3.159) becomes
y4-y2-w2=-2C. (3.161)

Figure 3.2

C=2
72 Non-linear ordinary differential equations

The graph of this family of curves is shown in Figure 3.3. IfO

both open and closed curves exist in the phase-plane, whereas if C < 0
or C ^ J only open curves exist. Hence, periodic solutions of the
equation exist only if the initial conditions are such that 0 ^ C < J, and
at t = 0 the point lies within the shaded region of Figure 3.3. An
example of this is shown for C = ^ . The phase-plane method is also of
importance in problems of stability of solutions of non-linear systems
of the form (3.155). The technique requires knowledge of the fixed or
critical points (y0, w0) of (3.155) defined by
Wo) = 0. (3.162)
The stability problem is not central to our discussion of obtaining
solutions of non-linear equations and, accordingly, we shall not pursue
it here.
For the Duffing equation, the differential relation (3.158) defining
the curves in the phase-plane is simple to integrate since the variables
may be separated. This may not be the case for other non-linear

Figure 3.3

c—i-
 Problems 3 73

equations. For example, the van der Pol equation

- l ) ^ + >>=0, (3.163)

where fi is a constant, has curves in the phase-plane defined by

dy w

This cannot be integrated by separation of variables but solution

curves can be sketched using, for example, the method of isoclines1.
For a more detailed account of analytic techniques for studying
non-linear differential equations, including stability problems and
chaotic behaviour, the reader is referred to a standard text*. In
general, however, non-linear equations require numerical methods for
their solution.

Problems 3
1. Obtain a solution of the equation

given y(0) = 0 and / ( 0 ) = \ll.

2. By writing dy/dx = p, solve

given y = dy/dx = 1 when x = 1.

3. By writing xy' — y = v(x), solve
x3y"={xy' -yf.
y
4. By writing t = z, solve

In x x In x
5. Show that the equation
y" = xy + 2y3 + \
f
E. Kreysig, Advanced Engineering Mathematics (Wiley, New York, 1988)
Section 1.10.
* D. W. Jordan and P. Smith, Nonlinear Ordinary Differential Equations,
(O.U.P, Oxford, 1987)
74 Non-linear ordinary differential equations

has a first integral y' = \x + y2. By finding the general solution of

this Riccati equation in terms of Airy functions, obtain a solution of
the original equation containing one arbitrary constant.
6. Verify that

+
' W2
where C is a constant, satisfies Pinney's equation
y"' + p(x)y = C/y3,
where u and v are independent solutions of
z" + p(x)z = Q
and W = uv' — vu'. (For equations with no first derivative term
dz/dx, W, the Wronskian, is a constant). Hence show that

is a solution of
y" + y = 2/y3.
7. Show that

(l)
J2/V3 V(2M 4 - 3M2 + 1) " V2 \ V2 ' 3 / '

where F and £ are elliptic integrals of the first and second kinds,
respectively.
8. Show that the inhomogeneous Duffing equation
y" + ay 4- by3 = A cos(3cot)
has an exact solution

y = \-g-j cos(o>o,
provided