Академический Документы
Профессиональный Документы
Культура Документы
a r t i c l e i n f o a b s t r a c t
Keywords: An eigenfunction approach to compute the particular solution of constant coefficient ordin-
Ordinary differential equations ary differential equations is presented. It is shown that the exponentials are the eigenfunc-
Constant coefficient tions of such equations. Solutions corresponding to products of powers and exponentials
Particular solution are obtained. The singular case is studied and a fast algorithm for its implementation is
Eigenfunction
presented.
Transfer function
Ó 2014 Elsevier Inc. All rights reserved.
1. Introduction
The constant coefficient ordinary differential equations are ubiquitous. Their mathematical manageability and practical
usefulness make them odd tools, even if they are not the best, for studying or modeling most natural or artificial systems.
This motivates their development since a long time ago and made them object of many books.
Perhaps we should not expect to find new features on such equations. This is a wrong thought as shown in [2,3] and moti-
vates us to propose a different approach for computing the particular solution for the special and of prime importance that is
the exponential case.
In engineering applications the exponentials or particularly the sinusoids are used to study the so-called steady state of
the filters, name usually given to the systems described by ordinary differential equations. Normally they are written in the
general format [5]
X
N X
M
ak Dak yðtÞ ¼ bk Dbk xðtÞ; ð1Þ
k¼0 k¼0
with t 2 R; the symbol D represents the derivative operator. The parameters ak and bk are the derivative orders. In the so-
called commensurate case we write ak ¼ bk ¼ ka. For now we will consider a equal to 1. The generalisation for fractional
orders will be presented in a future paper. On writing the second member with this format it is intended to consider practical
implementations where we need to consider derivatives on the input and, if studying the transient behaviour, the corre-
sponding initial conditions. As said above we will treat the steady state solution.
Particularising (1) for a ¼ 1 we have
X
N X
M
ak Dk yðtÞ ¼ bk Dk xðtÞ: ð2Þ
k¼0 k¼0
http://dx.doi.org/10.1016/j.amc.2014.01.036
0096-3003/Ó 2014 Elsevier Inc. All rights reserved.
M.D. Ortigueira / Applied Mathematics and Computation 232 (2014) 254–260 255
In current applications we assume that M 6 N for stability reasons. From a pure mathematical point of view, it is not nec-
essary to have such assumption provided that we work in the context of the generalised functions [10]. For wide classes of
functions we can use the Laplace transform (two-sided) or the Fourier transform [7] to find the particular solution of (2).
However the functions xðtÞ with the format
X
K
xðtÞ ¼ ebt dk t k ;
k¼0
used in [3] do not have neither Laplace transform nor Fourier transform, even in distributional sense. Jia & Sogabe treated
this subject by presenting an explicit formulation for the particular solution of equations of the type stated above, but they
put M ¼ 0. Other interesting approaches can be found in [1,2,4,8], but we think it is possible to get simpler algorithms. In the
following and without loosing generality, because Eq. (2) is linear, we will consider the case
Let xðtÞ and yðtÞ be two functions defined on R. The convolution is a binary operation defined with generality by:
Z 1
xðtÞ yðtÞ ¼ xðsÞyðt sÞds: ð4Þ
1
As it is easy to verify, the neutral element of the convolution is the Dirac impulse distribution, dðtÞ. Following a current pro-
cedure in Signal Processing, consider the solution, yðtÞ, of (2) when xðtÞ is an impulse, xðtÞ ¼ dðtÞ. We call it Impulse Response
and represent it by hðtÞ. We have then
X
N X
M
ak Dk hðtÞ ¼ bk Dk dðtÞ: ð5Þ
k¼0 k¼0
This means that the solution of (2) is the convolution of xðtÞ with the impulse response
yðtÞ ¼ hðtÞ xðtÞ ð6Þ
Theorem 1. Eigenfunction
The particular solution of the differential Eq. (2) when xðtÞ ¼ est s 2 C is given by
yðtÞ ¼ HðsÞest ; ð7Þ
provided that HðsÞ exists.
with
Z 1
HðsÞ ¼ hðsÞess ds; ð8Þ
1
we obtain (7). h
This theorem shows that the exponential defined on R is the eigenfunction of the constant coefficient ordinary differential
Eq. (2). The eigenvalue, HðsÞ, is the Transfer Function of the system defined by the differential Eq. (2) and is the Laplace trans-
form of the impulse response.
256 M.D. Ortigueira / Applied Mathematics and Computation 232 (2014) 254–260
In the following we will consider that the characteristic polynomial in the denominator is not zero for the particular value of s
at hand. Later we will consider the cases where the characteristic polynomial is zero (s is a pole).
It is interesting to remark here that in general there are at least two impulse responses corresponding to a given transfer
function. It depends on the choice of the associated region of convergence pffiffiffi[7]. To examplify,
pffiffiffi consider the transfer function
HðsÞ ¼ s3 þs214sþ2 of the Example 1, below. It has 3 poles at s ¼ 1; s ¼ 1 þ 3; s ¼ 1 3. Now pass a vertical straight line
by each pole. We define 4 regions (vertical strips) on the complex plane. Inverting HðsÞ with the Fourier–Mellin integral using
as integration path a straight line in a given region of convergence, we obtain an impulse response. For this case, there are 4
impulse responses having the same expression for the transform, HðsÞ, but different region of convergence. This is an advan-
tage of using the two sided Laplace transform. The one-sided, only gives one.
The sinusoidal case – In a particular setting, put s ¼ ix0 . We obtain immediately
Example 2. Consider again the above equation, but change the second member:
y000 þ y00 4y0 þ 2y ¼ x00 4x
and assume that xðtÞ ¼ eipt . Then
p2 þ 4
yðtÞ ¼ eipt :
ip þ p2 þ 4ip 2
3
From this result it would be immediate to compute the solution for xðtÞ ¼ cosðptÞ or xðtÞ ¼ sinðptÞ. Remark that if
xðtÞ ¼ ei2t ; yðtÞ ¼ 0.
To go further we are going to consider the case xðtÞ ¼ tK ebt . It is not difficult to see that we can write
K
d
xðtÞ ¼ lim K
est :
s!b ds
M.D. Ortigueira / Applied Mathematics and Computation 232 (2014) 254–260 257
Theorem 3. Power.exponential
The particular solution of the differential Eq. (2) when xðtÞ ¼ tK ebt is given by
K
d ½HðsÞest
yðtÞ ¼ lim K
; ð12Þ
s!b ds
Using the Leibniz rule we can obtain another expression for yðtÞ stated as follows.
Theorem 4. The particular solution of the differential Eq. (2) when xðtÞ ¼ tK ebt is given by:
XK
K
yðtÞ ¼ HðjÞ ðbÞt Kj ebt ; ð13Þ
j¼0
j
Example 3. Return back to the above example and as in [3] put xðtÞ ¼ 4t. We obtain immediately
X1
1 ðjÞ
yðtÞ ¼ H ð0Þttj :
j¼0
j
Consider now the situation where the characteristic polynomial in the denominator has an mth order root for s ¼ b. To look
for a solution return back to (7) and (9) to write
AðsÞyðtÞ ¼ BðsÞest :
When s goes to b the first member becomes 0:yðtÞ. So the solution is undetermined.
Assume for now that xðtÞ ¼ wðtÞebt . To obtain a solution of (2) we start by doing there the substitution
yðtÞ ¼ v ðtÞ:ebt : ð17Þ
Let D represent again the derivative operator. According to [2,3],
AðDÞ v ðtÞ:ebt ¼ AðD þ bÞv ðtÞ
and
BðDÞ wðtÞ:ebt ¼ BðD þ bÞwðtÞ:
With these changes we obtain a new differential equation with transfer function Hðs þ bÞ. In fact we moved the root of AðsÞ
from s ¼ b to s ¼ 0. This means that we have m null coefficients that we remove leading to a new equation
258 M.D. Ortigueira / Applied Mathematics and Computation 232 (2014) 254–260
X
Nm X
M
k Dkþm v ðtÞ ¼
a Dk wðtÞ;
b ð18Þ
k
k¼0 k¼0
where ak ; k ¼ 0; 1; . . . ; N m are the coefficients of the new characteristic polynomial AðsÞ ¼ AðsþbÞ
sm
and numerator polynomial
BðsÞ ¼ Bðs þ bÞ.
Now perform a new substitution uðtÞ ¼ Dm v ðtÞ to obtain
X
Nm X
M
k Dk uðtÞ ¼
a Dk wðtÞ:
b ð19Þ
k
k¼0 k¼0
For the particular case we are interested in, wðtÞ ¼ tK , we can use (14). Let D1 represent the anti-derivative.1 So
v ðtÞ ¼ Dm uðtÞ, allowing to obtain
Theorem 5. Singular case
The particular solution of the differential Eq. (2) when xðtÞ ¼ tK ebt with AðbÞ ¼ 0 is given by
" #
XK
bt m K ðjÞ Kj
yðtÞ ¼ e D H ð0Þt ; ð20Þ
j¼0
j
with
BðsÞ sm Bðs þ bÞ
HðsÞ ¼ ¼ :
AðsÞ Aðs þ bÞ
It is not difficult to show that (20) can be written
" #
XK
K ðK jÞ!
yðtÞ ¼ ebt HðjÞ ð0Þ t Kþmj ; ð21Þ
j¼0
j ðK þ m jÞ!
but in practice it is preferable to obtain first uðtÞ and then compute the anti-derivative term by term. As seen the net effect of
the singularity is an increase in the exponent of the power function in xðtÞ.
The implementation may be cumbersome even with transfer functions of low orders. So we need ways to make easy the
computations. We are going to present two.
1
D1 Df ðtÞ ¼ DD1 f ðtÞ ¼ f ðtÞ; [5] essentially the mth order primitive without primitivation constants.
2
A slightly different from the result showed in [3].
M.D. Ortigueira / Applied Mathematics and Computation 232 (2014) 254–260 259
allowing an easy computation of the new polynomial. This result can be obtained recursively.
2. Computation of the transfer function derivatives at s ¼ 0. To compute yðtÞ using (13), (20), and (21) we need the succes-
sive derivatives of HðsÞ at the origin. The direct computation although not difficult may be very boring and leading to very
large expressions which can originate errors. To avoid it is better to compute the MacLaurin series associated to HðsÞ. This
is not difficult to do because:
(a) HðsÞ can be decomposed in a sum of a polynomial and a proper fraction.
(b) The fractional part can be decomposed in a sum of partial fractions. This is easy to do if the poles are simple by the
residue theorem. If they are multiple, we can use instead of the residue theorem an algorithm presented in [9] that
does not need derivative computations.
(c) Each partial fraction has a MacLaurin series that we obtain using the geometric series.
This algorithm requires the knowledge of the poles. An alternative without such requirement can be found in [6]. Essen-
tially it consists of a decomposition of HðsÞ in a Laurent series that is inverted term by term to give a MacLaurin series.
s2 1
HðsÞ ¼ ¼
ðs þ 2Þ 4ðs þ 2Þ þ 3ðs þ 2Þ2 þ 4ðs þ 2Þ 4
4 3 s2 þ 4s þ 3
leading to
u0 ðtÞ ¼ 1=3;
1 4
u1 ðtÞ ¼ t ;
3 9
1 2 8 26
u2 ðtÞ ¼ t þ ;
3 9 27
and
1 3 4 2 26 80
u3 ðtÞ ¼ t t þ t :
3 3 9 27
Joining these solutions according to the right hand side of the initial equation, we have
1 3 20
uðtÞ ¼ t 2t2 þ 5t ;
3 3
leading to
1 5 1 4 5 3 10 2
v ðtÞ ¼ t t þ t t :
60 6 6 3
The final solution will be
260 M.D. Ortigueira / Applied Mathematics and Computation 232 (2014) 254–260
1 5 1 4 5 3 10 2 2t
yðtÞ ¼ t t þ t t e ;
60 6 6 3
in agreement with [3].
3. Conclusions
An eigenfunction approach to the computation of the particular solution of constant coefficient ordinary differential
equations was presented. We started by showing that the exponentials are the eigenfunctions of such equations. Solutions
corresponding to products of powers and exponentials were obtained. For both inputs we studied the singular case and
proposed a fast algorithm for its implementation.
Acknowledgments
This work was partially funded by National Funds through the Foundation for Science and Technology of Portugal.
References
[1] E. Deeba, S.A. Khuri, S. Xie, An algorithm for solving a nonlinear integro-differential equation, Appl. Math. Comput. 115 (2000) 123–131.
[2] R.C. Gupta, On particular solutions of linear difference equations with constant coefficients, SIAM Rev. 40 (1998) 680–684.
[3] J. Jia, T. Sogabe, On particular solution of ordinary differential equations with constant coefficients, Appl. Math. Comput. 219 (2013) 6761–6767.
[4] R.K. Nagle, E.B. Saff, Fundamentals of Differential Equations, Addison-Wesley, New York, 1996.
[5] M.D. Ortigueira, Fractional calculus for scientists and engineers, Lecture Notes in Electrical Engineering, vol. 84, Springer, Dordrecht, 2011.
[6] M.D. Ortigueira, J.J. Trujillo, V.I. Martynyuk, F.J.V. Coito, A simple transfer function inversion, in: Proceedings of the International Conference on
Mathematical Methods in Engineering, July 22–26, ISEP, Porto, Portugal, 2013.
[7] M.J. Roberts, Signals and Systems: Analysis Using Transform Methods and Matlab, McGraw-Hill, 2003.
[8] W. Wang, Z. Li, A mechanical algorithm for solving ordinary differential equation, Appl. Math. Comput. 172 (2006) 568–583.
[9] D. Westreich, Partial fraction expansion without derivative evaluation, IEEE Trans. Circuits Syst. 38 (6) (1991).
[10] A.H. Zemanian, Distribution Theory and Transform Analysis, Dover Publications, New York, 1987.