7

Multiple solutions of a nonlinear problem

It is well known that many nonlinear problems have multiple solutions. For
example, let us consider again the so-called Duffing oscillator in space, gov-
erned by
v  + (v − v 3 ) = 0, (7.1)
subject to the boundary conditions

v(0) = v(π) = 0, (7.2)

where the prime denotes the derivation with respect to ξ. In Chapter 6, we

use the homotopy analysis method to solve the same problem and correctly
discover its critical condition  = 1 for the simple bifurcation and express its
solution by such a set of base functions

{sin[(2m + 1)ξ] | m ≥ 0} . (7.3)

Notice that there exist an infinite number of sets of base functions denoted
by
{sin[(2m + 1)κ ξ] | m ≥ 0, κ ≥ 1} , (7.4)
where κ ≥ 1 is a positive integer, which can be used to express a real function
satisfying the boundary conditions (7.2). This implies that Equations (7.1)
and (7.2) might have multiple solutions. This is indeed true. We show in this
chapter that, using base functions denoted by (7.4), we can gain all multiple
solutions of Equations (7.1) and (7.2) by means of the homotopy analysis
method.
Without loss of any generality, define

A = v(π/2κ), v(ξ) = A u(ξ). (7.5)

Then, Equation (7.1) becomes

u + (u − A2 u3 ) = 0, u(0) = u(π) = 0. (7.6)

Note that A is unknown in the above equation. From (7.5), it holds

u(π/2κ) = 1. (7.7)

© 2004 CRC Press LLC

7.1 Homotopy analysis solution
7.1.1 Zero-order deformation equation
Using the base functions (7.4) and the boundary conditions u(0) = u(π) = 0
and considering the nonlinearity of Equation (7.6), we express the solution
u(ξ) in the form
+∞

u(ξ) = cm sin[(2m + 1)κξ], (7.8)
m=0

where cm is a coefficient. This provides us with the so-called rule of solution

expression.
Under the rule of solution expression denoted by (7.8) and using (7.7), it
is straightforward to choose

u0 (ξ) = sin(κ ξ) (7.9)

as an initial guess of u(ξ), where κ ≥ 1 is an integer. To obey the rule of

solution expression denoted by (7.8), we choose an auxiliary linear operator

∂ 2 Φ(ξ; q)
L[Φ(ξ; q)] = + κ2 Φ(ξ; q) (7.10)
∂ξ 2

such that
L[C1 sin(κ ξ) + C2 cos(κ ξ)] = 0, (7.11)
where C1 and C2 are coefficients. Furthermore, from Equation (7.6), we define
the nonlinear operator

∂ 2 Φ(ξ; q)  
N [Φ(ξ; q), α(q)] = 2
+  Φ(ξ; q) − α2 (q)Φ3 (ξ; q) , (7.12)
∂ξ

where q ∈ [0, 1] is the embedding parameter, Φ(ξ; q) is an unknown function

of ξ and q, α(q) is an unknown function dependent upon q. Let
= 0 denote
an auxiliary parameter and H(ξ) = 0 an auxiliary function. We construct the
so-called zero-order deformation equation

(1 − q)L[Φ(ξ; q) − u0 (ξ)] =
q H(ξ) N [Φ(ξ; q), α(q)], (7.13)

subject to the boundary conditions

Φ(0; q) = Φ(π; q) = 0. (7.14)

When q = 0, the solution of Equations (7.13) and (7.14) is

Φ(ξ; 0) = u0 (ξ), ξ ∈ [0, π]. (7.15)

© 2004 CRC Press LLC

When q = 1, Equations (7.13) and (7.14) are equivalent to Equations (7.6),
provided
Φ(ξ; 1) = u(ξ), α(1) = A. (7.16)
Thus, Φ(ξ; q) varies (or deforms) from the initial approximation u0 (ξ) =
sin(κξ) to the exact solution u(ξ) of Equations (7.6), as does α(q) from its
initial approximation A0 to the exact value A = u(π/2κ). Note that the zero-
order deformation equation (7.13) contains the auxiliary parameter
and the
auxiliary function H(ξ). Assume that
and H(ξ) are properly chosen so that
the zero-order deformation equations (7.13) and (7.14) have solutions for all
q ∈ [0, 1] and that the terms

1 ∂ m Φ(ξ; q) 1 dm α(q)
um (ξ) = , Am = (7.17)
m! ∂q m q=0 m! dq m q=0

exist for m ≥ 1. Then, by Taylor’s theorem and using (7.15), we can expand
Φ(ξ; q) and α(q) in power series of q as follows
+∞

Φ(ξ; q) = u0 (ξ) + um (ξ) q m , (7.18)
m=1
+∞

α(q) = A0 + Am q m . (7.19)
m=1

Furthermore, assuming that
and H(ξ) are so properly chosen that the power
series (7.18) and (7.19) are convergent at q = 1, we have from (7.16) the
solution series
+∞

u(ξ) = u0 (ξ) + um (ξ), (7.20)
m=1
+∞

A = A0 + Am . (7.21)
m=1

7.1.2 High-order deformation equation

For brevity, write

uk = {u0 (ξ), u1 (ξ), u2 (ξ), · · · , uk (ξ)} , Ak = {A0 , A1 , A2 , · · · , Ak } .

Differentiating the zero-order deformation equations (7.13) and (7.14) m times

with respect to q and then dividing them by m! and finally setting q = 0, we
have the high-order deformation equation

L [um (ξ) − χm um−1 (ξ)] =
H(ξ) Rm (um−1 , Am−1 ), (7.22)

© 2004 CRC Press LLC

subject to the boundary conditions
um (0) = um (π) = 0, (7.23)
where χm is defined by (2.42) and
1 ∂ m−1 N [Φ(ξ; q), α(q)]
Rm (um−1 , Am−1 ) =
(m − 1)! ∂q m−1 q=0
= um−1 (ξ) +  um−1 (ξ)
m−1
 n  ⎡m−1−n m−1−n−j
⎤
   
− Ai An−i ⎣ uj (ξ) ur (ξ)um−1−n−j−r (ξ)⎦ .
n=0 i=0 j=0 r=0

(7.24)
Note that both um (ξ) and Am−1 are unknown, but we have only one dif-
ferential equation for um (ξ). So, the problem is not closed and an additional
algebraic equation is needed to determine Am−1 . Assume that H(ξ) is prop-
erly chosen so that the right-hand side term of the high-order deformation
equation (7.22) can be expressed by
µm


H(ξ) Rm (um−1 , Am−1 ) = bm,n (Am−1 ) sin[(2n + 1)κ ξ], (7.25)
n=0

where bm,n (Am−1 ) is a coefficient and the positive integer µm depends upon
H(ξ) and m. According to the property (7.11) of L, when bm,0 (Am−1 ) = 0,
the solution of the mth-order deformation equation (7.22) contains the term
ξ sin(κ ξ),
which disobeys the rule of solution expression denoted by (7.8). To avoid this,
we had to enforce
bm,0 (Am−1 ) = 0, (7.26)
which provides us with an additional algebraic equation for Am−1 . In this way,
the problem is closed. Thereafter, it is easy to gain the solution of Equation
(7.22), say,
µm
 bm,n
um (ξ) = χm um−1 (ξ) + sin[(2n + 1)κ ξ]
n=1
[1 − (2n + 1)2 κ2 ]
+ C1 sin(κξ) + C2 cos(κξ), (7.27)
where C1 and C2 are coefficients. Under the rule of solution expression de-
noted by (7.8), C2 must be zero. Note that the coefficient C1 cannot be
determined by the boundary conditions (7.23), which is automatically satis-
fied when C2 = 0. However, from (7.7), it should hold
um (π/2κ) = 0, (7.28)

© 2004 CRC Press LLC

which uniquely determines C1 . In this way, we gain Am−1 and um (ξ) succes-
sively.
At the N th-order of approximation, we have
N

u(ξ) ≈ u0 (ξ) + um (ξ), (7.29)
m=1
N
 −1
A ≈ A0 + Am . (7.30)
m=1

7.1.3 Convergence theorem

THEOREM 7.1
If the solution series (7.20) and (7.21) are convergent, where uk (ξ) is gov-
erned by Equations (7.22) and (7.23) under the definitions (7.24) and (2.42),
they must be the exact solution of Equations (7.6).

Proof: If the solution series (7.20) is convergent, it is necessary that

lim um (ξ) = 0, ξ ∈ [0, π].
m−>+∞

From (7.10), (2.42), and (7.22) and using the above expression, we have
+∞


H(ξ) Rk (uk−1 , Ak−1 )
k=1
m

= lim L[uk (ξ) − χk uk−1 (ξ)]
m→+∞
k=1
 m


=L lim [uk (ξ) − χk uk−1 (ξ)]
m→+∞
k=1

=L lim um (ξ)
m→+∞
= 0.
Since
= 0 and H(ξ) = 0, the above expression gives
+∞

Rk (uk−1 , Ak−1 ) = 0.
k=1

Substituting (7.24) into the above expression and simplifying it, we have, due
to the convergence of the series (7.20) and (7.21), that
+∞  ⎧   +∞ 2 +∞ 3 ⎫
d2  ⎨ +∞   ⎬
u k (ξ) +  u k (ξ) − A m u k (ξ) = 0.
dξ 2 ⎩ ⎭
k=0 k=0 m=0 k=0

© 2004 CRC Press LLC

Using (7.9) and (7.23), it holds
+∞
 +∞

uk (0) = uk (π) = 0.
k=0 k=0

Thus, as long as the solution series (7.20) and (7.21) are convergent, they
must be the exact solution of Equations (7.6). This ends the proof.

7.2 Result analysis

According to Theorem 7.1, we need only to properly choose an auxiliary func-
tion H(ξ) and an auxiliary parameter
to ensure that the solution series
(7.20) and (7.21) converge. As pointed out in Chapter 6, the auxiliary func-
tion H(ξ) can be chosen in many different forms without disobeying the rule
of coefficient ergodicity. For the sake of simplicity, we choose here
H(ξ) = 1. (7.31)
Then, using (7.9) and (7.24), we have

H(ξ) R1 (u0 , A0 )
 
3 1
=
 − κ −  A0 sin(κ ξ) +
 A20 sin(3κ ξ),
2 2
(7.32)
4 4
which gives according to (7.25) that
 
2 3 2 1
b1,0 =
 − κ −  A0 , b1,1 =
 A20 .
4 4
Thus, from Equation (7.26), we have an algebraic equation
3
 − κ2 −  A20 = 0, (7.33)
4
which has the nonzero solution


2 κ2
A0 = ± √ 1− (7.34)
3 
when  > κ2 . Thus, for any a positive integer κ ≥ 1, the so-called bifurcation
occurs when
 = κ2 . (7.35)
This critical condition of bifurcations indicates that there exist multiple bi-
furcation points for large . Note that κ determines the set of base functions
denoted by (7.4). So, for large , there exist multiple solutions.

© 2004 CRC Press LLC

 Without the loss of generality, we consider here the two cases of κ = 2 and
κ = 3. Note that the convergence region and rate of the solution series (7.20)
and (7.21) are determined by the auxiliary parameter
. For a given  and a
positive integer κ, where  > κ2 ≥ 1, we can always find, by means of plotting
the so-called
-curves (see page 26 and §3.5.1) of A, a valid region of
to
ensure that the solution series (7.21) converges. For example, the
-curves of
A ∼
when κ = 2,  = 10, 25, 100, and κ = 3,  = 40, 90, 225 are as shown
in Figures 7.1 and 7.2, respectively. From these
-curves, it is clear that the
series (7.21) converges when  = 10 and κ = 2, 3 by means of
= −1, or
 = 40, κ = 2 and  = 90, κ = 3 by means of
= −1/2, or  = 100, κ = 2 and
 = 225, κ = 3 by means of
= −1/5. This is indeed true, as shown in Tables
7.1 and 7.2.
From Figures 7.1 and 7.2, the so-called valid region of
for A decreases
as  increases for a given κ. So, the absolute value of
should decrease
as  increases. It is found that, for any a given  and a given κ satisfying
 > κ2 ≥ 1, the series (7.21) is always convergent in the region

κ2 ≤  < +∞,

when %  &−1

=− 1+ 2 . (7.36)
3κ
Besides, the corresponding 10th-order approximation


%  &−10 κ2  2
A ≈ ± 1+ 2 1− 1.1803 + 3.9075 2 + 5.8128 4
3κ  κ κ
3 4 5 6
   
+ 5.1149 6 + 2.9466 8 + 1.1603 1 + 0.31602 12
κ κ κ 0 κ
7 8 9
+ 5.8726 × 10−2 14 + 7.1298 × 10−3 16 + 5.1396 × 10−4 18
κ  κ κ
10

+ 1.7001 × 10−5 20 (7.37)
κ

agrees well in the whole region κ2 ≤  < +∞ with the exact analytic result
given by the implicit formula
 
8κ2 A2
= 2 K , (7.38)
π (2 − A2 ) 2 − A2

where K denotes the complete elliptic integral of the first kind, as shown in
Figure 7.3. In fact, Figure 7.3 provides us with a complete bifurcation diagram
of the so-called Duffing oscillator in space problem.
Using the so-called homotopy-Padé technique (see page 38 and §3.5.2),
we can greatly accelerate the convergence of the series (7.21), as shown in
Tables 7.3 and 7.4. It is found that the [m, m] homotopy-Padé approximant

© 2004 CRC Press LLC

does not depend upon the auxiliary parameter
. The [4, 4] homotopy-Padé
approximant

√ κ2 P ()
A≈2 3 1− (7.39)
 Q()
gives an accurate approximation of A in the whole region 1 ≤ /κ2 < +∞,
where
%  &
P () = 8665210296046039923 + 2500964782519057396
κ2
%  &2 %  &3
+ 604034298653768562 + 62408285303687028
κ2 κ2
%  &4
+ 3874319809940915 ,
κ2 %  &
Q() = 25430938337575455089 + 7921677254280814588
κ2
%  &2 %  &3
+ 1930521704826790758 2
+ 213027971364041596
κ κ2
%  &4
+ 13310678950379441 .
κ2
It is found that, as long as the series (7.21) of A is convergent, the corre-
sponding series (7.20) of u(ξ) given by the same value of
converges in the
whole region ξ ∈ [0, π], as shown in Figures 7.4 and 7.5. Due to the odd
nonlinearity of Equation (7.6), if u(ξ) is a solution, −u(ξ) must be also a
solution. However, for brevity, we do not give this kind of solution in Fig-
ures 7.4 and 7.5. The nonlinear problem of the so-called Duffing oscillator
in space has multiple solutions for large . For example, when  = 10, there
exist two nonzero solutions corresponding to κ = 1, two nonzero solutions to
κ = 2, and two nonzero solutions to κ = 3, respectively, so that there are
six nonzero solutions. In general, for any √ given  ≥ 1, the problem of the
so-called Duffing oscillator in space has 2[ ] nonzero solutions, where [x] de-
notes the integer part of x. Therefore, the larger the value of , the more the
multiple solutions, as shown in Figure 7.3. As  tends to infinity, there exists
an infinite number of solutions. Therefore, the nonlinear equation (7.1) with
boundary conditions (7.2) contains rather rich mathematical structure and a
complicated bifurcation diagram.
The rule of solution expression plays an important role in finding these
multiple solutions. This example clearly indicates that, by means of different
base functions, we can employ the homotopy analysis method to gain all
multiple solutions of some nonlinear problems. Indeed, the so-called rule of
solution expression of the homotopy analysis method provides us with a new
viewpoint and a different starting point to investigate nonlinear problems.

© 2004 CRC Press LLC

TABLE 7.1
The analytic approximations of A when  = 10 and κ = 2, 3 by means
of
= −1 and H(ξ) = 1.
Order of approximation κ=2 κ=3
2 0.8694142054 0.3643175731
4 0.8696932532 0.3643100899
6 0.8696857656 0.3643100194
8 0.8696860265 0.3643100187
10 0.8696860164 0.3643100187
12 0.8696860168 0.3643100187
14 0.8696860168 0.3643100187
16 0.8696860168 0.3643100187
18 0.8696860168 0.3643100187
20 0.8696860168 0.3643100187

TABLE 7.2
The analytic approximations of A when  = 40, κ = 2 and  = 90, κ = 3
by means of
= −1/2 and H(ξ) = 1.
Order of approximation  = 40, κ = 2  = 90, κ = 3
2 0.98070 0.98171
4 0.99912 0.99854
6 0.99613 0.99639
8 0.99634 0.99625
10 0.99656 0.99658
12 0.99635 0.99635
14 0.99649 0.99648
16 0.99641 0.99642
18 0.99645 0.99644
20 0.99643 0.99644
22 0.99644 0.99644
24 0.99644 0.99644
26 0.99644 0.99644
28 0.99644 0.99644
30 0.99644 0.99644

© 2004 CRC Press LLC

 TABLE 7.3
The [m, m] homotopy-Padé approximant of A when  = 10 and
κ = 2, 3 by means of H(ξ) = 1.
[m, m] κ=2 κ=3
[1, 1] 0.8694029457 0.3643104636
[2, 2] 0.8696902377 0.3643100178
[3, 3] 0.8696859569 0.3643100187
[4, 4] 0.8696860176 0.3643100187
[5, 5] 0.8696860168 0.3643100187
[6, 6] 0.8696860168 0.3643100187
[7, 7] 0.8696860168 0.3643100187
[8, 8] 0.8696860168 0.3643100187
[9, 9] 0.8696860168 0.3643100187
[10, 10] 0.8696860168 0.3643100187

TABLE 7.4
The [m, m] homotopy-Padé approximant of A when  = 40, κ = 2
and  = 90, κ = 3 by means of H(ξ) = 1.
[m, m]  = 40, κ = 2  = 90, κ = 3
[1, 1] 0.9747449855 0.9753179745
[2, 2] 0.9988803766 0.9988250895
[3, 3] 0.9960551840 0.9960761350
[4, 4] 0.9964957766 0.9964921829
[5, 5] 0.9964304420 0.9964305571
[6, 6] 0.9964370766 0.9964368336
[7, 7] 0.9964352860 0.9964352709
[8, 8] 0.9964353707 0.9964353614
[9, 9] 0.9964353314 0.9964353314
[10, 10] 0.9964353352 0.9964353355
[11, 11] 0.9964353351 0.9964353363
[12, 12] 0.9964353362 0.9964353363
[13, 13] 0.9964353363 0.9964353363
[14, 14] 0.9964353363 0.9964353363
[15, 15] 0.9964353363 0.9964353363

© 2004 CRC Press LLC

 2
k = 2

e = 10
A

e = 40

-1

e = 100

-2
-2.5 -2 -1.5 -1 -0.5 0
h-

FIGURE 7.1
The
-curves of A when κ = 2 and  = 10, 40, 100 by means of H(ξ) = 1.
Dash-dot-dotted line: 20th-order approximation of A when  = 10; dash-
dotted line: 20th-order approximation of A when  = 40; solid line: 20th-order
approximation of A when  = 100.

© 2004 CRC Press LLC

 2

k=3

1.5 e = 225

e = 10

e = 90
A

0.5

0
-3 -2 -1 0
h-

FIGURE 7.2
The
-curves of A when κ = 3 and  = 10, 90, 225 by means of H(ξ) = 1.
Dash-dot-dotted line: 20th-order approximation of A when  = 10; dash-
dotted line: 20th-order approximation of A when  = 90; solid line: 20th-order
approximation of A when  = 225.

© 2004 CRC Press LLC

 1.25

1 1

2
0.75

3
0.5
4
0.25 5 6
A

-0.25

-0.5

-0.75

-1

-1.25
0 10 20 30 40

FIGURE 7.3
The comparison of the 10th-order approximation (7.37) of A with the exact
implicit solution (7.38). Symbols: exact result; curve 1: κ = 1; curve 2: κ = 2;
curve 3: κ = 3; curve 4: κ = 4; curve 5: κ = 5; curve 6: κ = 6.

© 2004 CRC Press LLC

 1.25
k=2
1

0.75

0.5

0.25
v(x)

-0.25

-0.5

-0.75

-1

-1.25
0 1 2 3

FIGURE 7.4
The convergent analytic result of v(ξ) = Au(ξ) when κ = 2 and  = 10, 40, 100
by means of H(ξ) = 1. Dashed line: fifth-order approximation of v(ξ) when
 = 10 by means of
= −1; dash-dotted line: 10th-order approximation of
v(ξ) when  = 40 by means of
= −1/2; solid line: 20th-order approximation
of v(ξ) when  = 100 by means of
= −1/5.

© 2004 CRC Press LLC

 1.25 k = 3
1

0.75

0.5

0.25
v(x)

-0.25

-0.5

-0.75

-1

-1.25
0 1 2 3

FIGURE 7.5
The convergent analytic result of v(ξ) = Au(ξ) when κ = 3 and  = 10, 90, 225
by means of H(ξ) = 1. Dashed line: fifth-order approximation of v(ξ) when
 = 10 by means of
= −1; dash-dotted line: 10th-order approximation of
v(ξ) when  = 90 by means of
= −1/2; solid line: 20th-order approximation
of v(ξ) when  = 225 by means of
= −1/5.

© 2004 CRC Press LLC