This work is licensed under the Creative Commons Attribution 4.0 International License.
Introduction
Non-linear dynamic problems have fascinated applied mathematicians, physicists and engineers from a long time. Over the past few decades, applications in solid and structural mechanics as well as fluid mechanics have appeared, and currently, there is widespread interest in non-linear oscillators, strange attractors, as well as the chaotic and dynamical systems theories in the engineering and applied science communities.
Physical and mechanical oscillatory systems are often governed by non-linear differential equations. Unfortunately, with the exception of a few particular cases, the exact analytical solutions of such equations cannot be determined. In many cases, it is possible to replace the non-linear differential equation by a corresponding linear differential equation that approximates the original non-linear equation closely, to give useful results. Often such linearisation is not feasible or possible, and for this situation, the original differential equation itself must be directly dealt with.
However, in many cases, it is possible to compute accurate approximate analytical solutions of the equations. A large number of approximate methods, such as perturbation [1, 2, 3, 4, 5, 6], harmonic balance (HB) [7, 8, 9, 10, 11, 12, 13], homotopy perturbation [14], iteration [15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28] and so on, are commonly used for solving non-linear oscillatory systems. The perturbation method is mainly used for small non-linear problems. On the other hand, HB and iteration methods are mostly used for strong and small non-linear problems.
One important class of non-linear oscillators is conservative oscillators, in which the restoring force is not dependent on time, the total energy is constant and any oscillation is stationary. Despite the great elegance and simplicity of such equations, the solutions of specific problems are significantly hard to derive. Finding innovative methods to analyse and solve these equations has become an interesting subject in the field of ordinary and partial differential equations and dynamical systems. Using non-linear equations for most real-life problems is not always possible, and sometimes, it is not even advantageous to express exact solutions of non-linear differential equations explicitly in terms of elementary functions or independent spatial and/or temporal variables; however, it is possible to find approximate solutions.
Perturbation means grossly small change; so, the method is adopted when the non-linearity is small. Thus, in case of strong non-linearities, the perturbation method is not generally adopted. It is used to construct a uniformly valid periodic solution to second-order non-linear differential equations. A critical feature of the technique is a middle step that breaks the problem into ‘solvable’ and ‘perturbation’ parts. Perturbation theory is applicable if the problem at hand cannot be solved exactly but can be formulated by adding a ‘small’ term to the mathematical description of the exactly solvable problem.
The HB method is a procedure for determining analytical approximations to the periodic solutions of differential equations by using a truncated Fourier series representation. An important advantage of the method is that it can be applied to non-linear oscillatory problems for which the non-linear terms are not ‘small’, i.e. no perturbation parameter need to exist. A disadvantage of the method is that it is difficult to predict, for a given non-linear differential equation, whether a first-order HB calculation will provide a sufficiently accurate approximation to the periodic solution.
The iterative technique is particularly useful for calculating approximate periodic solutions and the corresponding frequencies of truly non-linear oscillators for small and large amplitudes of oscillation.
The main intention of this research is to investigate the approximate analytical solutions using the modified extended iterative method to decompose the secular term, so that the solution can be obtained by the iterative procedure. This means that we can use the extended iterative method to investigate many non-linear problems. The main thrust of this technique is that the obtained solution rapidly converges to the exact solutions.
Methodology
An extended iterative method is used to obtain the analytical solution of the quadratic non-linear oscillator. The procedure may be briefly described as follows.
A non-linear oscillator is modelled by the following expression:
\ddot x + f{\kern 1pt} (x) = 0,{\kern 1pt} {\kern 1pt} x{\kern 1pt} (0) = A,{\kern 1pt} {\kern 1pt} \dot x{\kern 1pt} (0) = 0{\kern 1pt},
where the overdots denote differentiation with respect to time t.
First, we choose the natural frequency Ω of this system. Then, adding Ω2x on both sides of Eq. (1), we obtain
\ddot x + {\Omega ^{\rm{2}}}{\kern 1pt} x = {\Omega ^{\rm{2}}}{\kern 1pt} x - f(x) \equiv G(x,{\kern 1pt} \Omega).
The extended iterative scheme is
{\ddot x_{k + 1}} + \Omega _k^2{\kern 1pt} {x_{k + 1}} = G({x_{k - 1}},\Omega) + {G_x}({x_{k - 1}},\Omega)({x_k} - {x_{k - 1}});{\kern 1pt} {\kern 1pt} {\kern 1pt} k = 1,2,...
where
{G_x} = {{\partial {\kern 1pt} G} \over {\partial {\kern 1pt} x}}
.
The right-hand side of Eq. (3) is essentially the first term in a Taylor series expansion of the function
G({x_k},{\dot x_k})
at the point
({x_{k - 1}},{\dot x_{k - 1}})
[29].
We have the direct iteration scheme of Eq. (2), as shown in Eq. (4):
{\ddot x_{k + 1}} + \Omega _k^2{x_{k + 1}} = G({x_k},{\Omega _k});{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} k = 0,{\kern 1pt} 1,{\kern 1pt} 2,{\kern 1pt}...
and xk+1 satisfies the condition
{x_{k + 1}}(0) = A.
The initial estimate is considered to be the following [10]:
{x_0}(t) = A{\kern 1pt} \cos {\kern 1pt} ({\Omega _0}{\kern 1pt} t).
The above procedure gives the sequence of following solutions: x1(t), x2(t), x3(t),.... The method can be extrapolated to any order of approximation; but due to growing algebraic complexity, the solution is confined to a lower order, usually the second [15].
Solution procedure
Let us consider the following non-linear inverse oscillator:
\ddot x + {x^2}{\kern 1pt} = {\kern 1pt} {\kern 1pt} 0
Adding Ω2x on both sides of Eq. (7), we get
\ddot x + {\Omega ^{\rm{2}}}x = {\Omega ^{\rm{2}}}x - {x^2} = G(x,\Omega)
where G(x,Ω) = Ω2x − x2 and Gx(x,Ω) = Ω2 − 2x.
According to Eq. (4), the direct iterative scheme of Eq. (8) is
{\ddot x_{k + 1}} + \Omega _k^2{x_{k + 1}} = \Omega _k^2{x_k} - x_k^2
The first approximation x1(t) and the frequency Ω0 are obtained by substituting k = 0 in Eq. (9), and using Eq. (6), we get
{\ddot x_1} + \Omega _0^2{\kern 1pt} {x_1} = \Omega _0^2{\kern 1pt} {x_0} - x_0^2{\kern 1pt}
where x0(t) = A cos(Ω0t) = A cos θ and θ = Ω0t.
According to Eq. (3), the extended iterative scheme of Eq. (8) is
\matrix{{{\kern 1pt} {{\ddot x}_{k + 1}} + \Omega _k^2{\kern 1pt} {x_{k + 1}}} \hfill & {= ({\Omega ^2}{\kern 1pt} {x_k} - x_k^2) + ({\Omega ^{\rm{2}}} - 2{x_k}){\kern 1pt} ({x_k} - {x_{k - 1}}).} \hfill \cr {\kern 1pt} \hfill & {= x_k^2 + {\Omega ^2}{x_{k - 1}} - 2{x_k}{x_{k - 1}}} \hfill \cr}
The second approximation x2(t) and the frequency Ω1 are obtained by substituting k = 1 in Eq. (17), and using Eq. (7), we get Eq. (18):
{\ddot x_2} + \Omega _1^2{\kern 1pt} {x_2} = x_1^2{\kern 1pt} + {\Omega ^2}{x_0} - 2{x_1}{x_0}
where x0(t) and x1(t) are given by Eqs (6) and (16).
Thus, Ω0 , Ω1 , Ω2 , respectively obtained from Eqs (12), (20) and (26), represent the approximation of the frequencies, and x1(t), x2(t), x3(t), respectively obtained from Eqs (16), (23) and (29), represent the corresponding approximate solutions of the oscillator represented by Eq. (7).
Another example
Let us consider the non-linear oscillator
\ddot x + x = - x{\dot x^2}
with initial condition x(x) = A,
\dot x(0) = 0
.
Obviously, Eq. (30) can be written as
\ddot x + {\Omega ^2}x = {\Omega ^2}x - x{\dot x^2} - x = G(x,\dot x)\;\;\;{\rm{Say}}
According to Eq. (4), the direct iterative scheme of Eq. (30) is
{\ddot x_{k + 1}} + \Omega _k^2{x_{k + 1}} = \Omega _k^2{x_k} - {x_k}\dot x_k^2
The first approximation x1(t) and corresponding frequency Ω0 are obtained from Eq. (32) by substituting k = 0:
{x_1} = \left({A - {1 \over {32}}(- 4 + {A^2}){a_{1,3}}} \right)\cos \theta + {1 \over {32}}\left({- 4 + {A^2}} \right){a_{1,3}}\cos 3\theta
where
{A_{1,3}} = {1 \over 4}{A^3}\Omega _0^2
and
{\Omega _0} = {2 \over {\sqrt {4 - {A^2}}}}.
Now, the extended iteration scheme, according to Eq. (3), is as follows:
{\ddot x_{k + 1}} + \Omega _k^2{x_{k + 1}} = G({x_{k - 1}},{\dot x_{k - 1}}) + {G_x}({x_{k - 1}},{\dot x_{k - 1}})({x_k} - {x_{k - 1}}) + {G_x}({x_{k - 1}},{\dot x_{k - 1}})({\dot x_k} - {\dot x_{k - 1}});\;k = 1,2, \ldots.
The second approximation x2(t) and the corresponding frequency Ω1 are obtained from Eq. (36) by substituting k = 1:
{y_2} = \left({A + {1 \over {\Omega _1^2}}\left({{{{a_{23}}} \over 8} + {{{a_{25}}} \over {24}}} \right)} \right)\cos \theta - {1 \over {\Omega _1^2}}\left({{{{a_{23}}} \over 8}\cos 3\theta + {{{a_{25}}} \over {24}}\cos 5\theta} \right)
where
\eqalign{& {a_{23}} = {{7{A^3}} \over {2(4 - A62)(- 4 + {A^2})}} - {{9{A^5}} \over {8(4 - {A^2})(- 4 + {A^2})}} + {{{A^7}} \over {16(4 - {A^2})(- 4 + {A^2})}} \cr & \,\,\,\,\,\,\,\,\,\,\,\,\, - {{4{A^3}{\Omega _1}} \over {\sqrt {4 - {A^2}} (- 4 + {A^2})}} + {{7{A^5}{\Omega _1}} \over {8\sqrt {4 - {A^2}} (- 4 + {A^2})}} + {{{A^7}{\Omega _1}} \over {32\sqrt {4 - {A^2}} (- 4 + {A^2})}} \cr & \,\,\,\,\,\,\,\,\,\,\,\,\,{{{A^3}\Omega _1^2} \over {2(4 - {A^2})(- 4 + {A^2})}} - {{{A^5}\Omega _1^2} \over {4(4 - {A^2})(- 4 + {A^2})}} + {{{A^7}\Omega _1^2} \over {32(4 - {A^2})(- 4 + {A^2})}} \cr}{a_{25}} = {{{A^5}} \over {8(4 - {A^2})(- 4 + {A^2})}} - {{{A^7}} \over {32(4 - {A^2})(- 4 + {A^2})}} + {{3{A^5}{\Omega _1}} \over {8\sqrt {4 - {A^2}} (- 4 + {A^2})}} - {{3{A^7}{\Omega _1}} \over {32\sqrt {4 - {A^2}} (- 4 + {A^2})}}\eqalign{& {\Omega _1} = (- 128{A^2} + 40{A^4} - 2{A^6} - 2(4096{A^4} - 2560{A^6} + 528{A^8} - 40{A^{10}} + {A^{12}} + 16384(4 - {A^2}) \cr & \,\,\,\,\,\,\,\,\,\,\,\,\, - 11264{A^2}(4 - {A^2}) + 1680{A^4}(4 - {A^2}) + 32{A^6}(4 - {A^2}) - {A^8}(4 - {A^2}){)^{1/2}})/(2(- 128\sqrt {4 - {A^2}} \cr & \,\,\,\,\,\,\,\,\,\,\,\,\, + 2{A^2}\sqrt {4 - {A^2}} + {A^4}\sqrt {4 - {A^2}})) \cr}
In a similar way, the method can be used for higher-order approximations.
Results and discussion
An iterative approach to obtain the approximate solution of the ‘quadratic non-linear oscillator’ is presented. The presented technique is very simple for solving algebraic equations analytically, and the approach is different from other existing approaches for adopting truncated Fourier series. This process significantly improves the results.
Here, the first, second and third approximate frequencies Ω0,Ω1 and Ω2 have been calculated, and the results are given in Table 1.
Adopted approximate frequencies of
\ddot x + {x^2} = 0
.
Exact frequency
{\Omega _e} = 0.914681\sqrt A
Amplitude A
First approximate frequencies, Ω0
0.921318\sqrt A
Second approximate frequencies, Ω1
0.914752\sqrt A
Third approximate frequencies, Ω2
0.91467\sqrt A
Error (%)
0.73
Error (%)
0.0078
Error (%)
0.0012
To compare the approximate frequencies, we have also given the existing results determined by Mickens and Ramadhani [9], Belendez et al. [14], Hosen [13] and Haque and Hossain [24], which are shown in Table 2. Fortunately, this current method gives significantly better results than the other formulae.
Comparison of the approximate frequencies with exact frequency Ωe of
\ddot x + {x^2} = 0
.
To show the accuracy, the percentage of errors is calculated by the following definition:
{\rm{Error}} = \left| {{{{\Omega _e} - {\Omega _k}} \over {{\Omega _e}}}} \right| \times 100\%
where Ωk(k = 0,1,2,...) represents the approximate frequencies obtained by the present method, and Ωe represents the corresponding exact frequency of the oscillator.
Though the method has been illustrated for a quadratic non-linear oscillator, it is also valid for other oscillators. The same technique has been applied to solve the non-linear jerk oscillator. Herein, we have calculated the approximate frequencies Ω0, Ω1 and Ω2, respectively, in Section 4. The results are given in Table 3 for five particular values of A; to compare the approximate frequencies, we have also given the existing results determined by Gottlieb [?]. We see that the presented technique gives us emphatically better results than the Gottlieb [30] technique.
Comparison of the approximate periods with exact periods Te of
\ddot x + x = - x{\dot x^2}
.
A
Te
T Er(%)
TG Er(%)
0.1
6.275334
6.275333 3.43 e−6
6.2753264 1.21 e−4
0.2
6.251809
6.251809 3.61 e−7
6.251690 1.90 e−3
0.5
6.088449
6.088449 3.01 e−6
6.083668 7.85 e−2
1
5.527200
5.527434 4.25 e−3
5.441398 1.55
1.5
4.690247
4.709049 4.7 e−1
4.155936 11.39
T denotes the modified approximate period and TG denotes approximate period obtained by Gottlieb [30]. Er(%) denotes percentage error.
Convergence and consistency analysis
We know that the basic idea of iterative methods is to construct a sequence of solutions xk (as well as frequencies Ωk) that have the property of convergence:
{x_e} = \mathop {\lim}\limits_{k \to \infty} {x_k};\;{\rm{or}}\;,{\Omega _e} = \mathop {\lim}\limits_{k \to \infty} {\Omega _k}.
Here, xe is the exact solution of the given non-linear oscillator.
In the present method, the solution yields less error in each iterative step compared to the previous iterative step, and finally, |Ω2 − Ωe| = |0.91467 − 0.914681 | < ɛ, where ɛ is a small positive number and A is chosen to be unity. From this, it is clear that the adopted method is convergent.
An iterative method of the form represented by Eq. (4), with initial estimate given in Eq. (5) is said to be consistent if
\mathop {\lim}\limits_{k \to \infty} |{x_k} - {x_e}| = 0,\;{\rm{or}}\;,\mathop {\lim}\limits_{k \to \infty} |{\Omega _k} - {\Omega _e}| = 0
In the present analysis, we see that
\mathop {\lim}\limits_{k \to \infty} |{\Omega _k} - {\Omega _e}| = 0,\;{\rm{as}}\;|{\Omega _2} - {\Omega _e}| = 0.
Thus, the consistency of the method is achieved.
Conclusion
It is noted that Mickens and Ramadhani [9] found only the second approximate frequencies by the HB method. Belendez et al. [14] found up to the third approximate frequencies by using a modified He's homotopy perturbation method. Again, Hosen [13] found up to the third approximate frequencies by using a modified HB method; Haque and Hossain [24] found up to the fourth approximate frequencies by the iteration method.
In our study, it is seen that the third-order approximate frequency obtained by the adopted method is almost same as the exact frequency. It is found that, in most of the cases, our solution gives significantly better results than other existing results. The advantages of this method include its simplicity and computational efficiency.