1 views

Uploaded by Matija Korvin

A Novel Class of Fractionally Orthogonal Quasipolynomials and New Fractional Quadrature Formulas

- PP & QM Integration
- Article Methode Wilson Tetha
- dfht kim mmm
- Numerical integration 1d element
- area under a curve - rectangle format 12-9-2017
- Final Assignment(Part 1)
- sem1
- NumericalMethodsT264UnitIVByDrNVNagendram.pdf
- numerical technique for guide
- Lecture 3 4
- ISME 2008-A Truly Meshless Element Free Galerkin Method by a General Approach for Evaluation of Domain Integrals
- abs_integrate.pdf
- fem gauss
- Theory of the Monte Carlo Method for Semiconductor
- Meshless Computation
- 1-2
- Métodos Numéricos
- 1-IN-GATE18_schedule.pdf
- PF
- UT Dallas Syllabus for math2414.001.10f taught by Bentley Garrett (btg032000)

You are on page 1of 14

journal homepage: www.elsevier.com/locate/amc

and new fractional quadrature formulas

Milan R. Rapaić a,⇑, Tomislav B. Šekara b, Vidan Govedarica c

a

Faculty of Technical Sciences, University of Novi Sad, Serbia

b

Faculty of Electrical Engineering, University of Belgrade, Serbia

c

Faculty of Electrical Engineering, University of Istočno Sarajevo, Bosnia and Herzegovina

a r t i c l e i n f o a b s t r a c t

Keywords: A novel class of quasi-polynomials orthogonal with respect to the fractional integration

Fractional quadratures operator has been developed in this paper. The related Gaussian quadrature formulas for

Orthogonal quasi-polynomials numerical evaluation of fractional order integrals have also been proposed. By allowing

Gaussian quadratures the commensurate order of quasi-polynomials to vary independently of the integration

Hermitean quadratures

order, a family of fractional quadrature formulas has been developed for each fractional

Interpolation

Fractional calculus

integration order, including novel quadrature formulas for numerical approximation of

classical, integer order integrals. A distinct feature of the proposed quadratures is high

computational efﬁciency and ﬂexibility, as will be demonstrated in the paper. As auxiliary

results, the paper also presents methods for Lagrangian and Hermitean quasi-polynomial

interpolation and Hermitean fractional quadratures. The development is illustrated by

numerical examples.

Ó 2014 Elsevier Inc. All rights reserved.

1. Introduction

In classical numerical integration theory [1,2] the primary objective is to design a quadrature formula

Z b X

n

f ðsÞgðsÞds Ai f ðti Þ ð1Þ

a i¼0

equipped with a desired set of properties. In the current paper, both the integrand f and the weight function g are assumed

to be real-valued function of real argument, i.e. f : ða; bÞ ! R and g : ða; bÞ ! R. The weight function g is usually assumed to

be non-negative. We refer to the points t i at which the integrand f is evaluated as grid points. Without loss of generality we

assume that a 6 t 0 < < tn 6 b. Real numbers Ai are called quadrature coefﬁcients. Quadrature formula (1) using n þ 1

grid points is said to be of order n.

Several groups of quadrature formulas can be distinguished. Among the most widely utilized ones are those of Newton–

Cotes and Gauss types. Newton–Cotes quadratures use ﬁxed and pre-speciﬁed grid values t i . Consequently, a Newton–Cotes

formula of order n has exactly n þ 1 adjustable parameters (Ai ), and can be made accurate for integrands equal to polynomial

functions of order up to n. Gaussian quadratures, on the other hand, allow for free selection of both grid points ti and

⇑ Corresponding author.

E-mail addresses: rapaja@uns.ac.rs (M.R. Rapaić), tomi@etf.rs (T.B. Šekara), vidangov@yahoo.com (V. Govedarica).

http://dx.doi.org/10.1016/j.amc.2014.07.084

0096-3003/Ó 2014 Elsevier Inc. All rights reserved.

M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219 207

quadrature coefﬁcients Ai . Thus, all 2ðn þ 1Þ parameters are chosen freely and, consequently, a Gaussian integration formula

can be made exact for all polynomial integrands of order 2n þ 1 or less.

The simultaneous evaluation of both t i and Ai requires solving a systems of non-linear equations. This system is not easily

solvable in most cases. However, it turns out that the grid points ti can be chosen as zeros of polynomials belonging to suit-

ably chosen orthogonal sets. Once the evaluation points are determined, the quadrature coefﬁcients are easily found. For fur-

ther details regarding orthogonal polynomials and classical Gaussian quadrature formulas we refer to [2,3].

Fractional calculus is mathematical theory of non-integer order differentiation and integration. In recent decades frac-

tional calculus is increasingly utilized in description of different natural phenomena, as well as in solving various engineer-

ing problems. A detailed survey of numerous ﬁelds of application of fractional calculus is beyond the scope of the present

paper. A recent accounts can be found in [4–7]. The primary operator of fractional calculus is fractional integral of order

a>0

Z t

1

a

0 It f ¼ f ðsÞðt sÞa1 ds; ð2Þ

CðaÞ 0

where C stands for the well-known Euler’s Gamma function. This operator is known as the (Left) Riemann–Liouville frac-

tional integral of order a. For integer values of a (a 2 N) formula (2) reduces to the well-known Cauchy repeated integration

formula. Thus, for integer values of the integration order the fractional integral becomes the classical, repeated or n-fold inte-

gral. When a tends to zero, 0 Iat reduces to the identity operator (the limit, however, must be taken in the weak, distributional

sense). We refer to [8–11] for further details on fractional calculus, alternative deﬁnitions of fractional integral and differ-

ential operators, and related concepts.

Numerous methods have been proposed for numerical evaluation of fractional integrals. A method based on ﬁnite mem-

ory approximation and Grünwald–Letnikov deﬁnition has been presented in [8]. A method based on discrete approximation

of (2) is presented in [12], with similar methods generalizing and extending classical Trapezoidal rule proposed in [13–15].

Some novel numerical approximation methods based on Haar wavelet approximation theory are presented in [16,17]. Efﬁ-

cient numerical methods for fractional differential equations have also been investigated in [18], with related numerical

issues considered in [19]. For recent methods for evaluation of fractional derivatives we refer to [20,21].

A plethora of methods for practical realization of fractional order systems has been developed. Each of these methods can

be seen as an approximation of the inﬁnite dimensional fractional order system by a suitable ﬁnite dimensional ‘‘rational’’

approximation, i.e. in a form of FIR or IIR ﬁlters of sufﬁciently high order. A survey of analog approximations is presented in

[22]. Examples of digital approximations can be found in [23,24]. Optimal rational approximations of fractional order sys-

tems, and fractional integrators in general, with special emphasis on retaining accurate steady state gain are proposed in

[25]. Rational approximations by means of fractional Laguerre basis have been discussed in [26,27].

Classical quadrature formulas assume that the integrand is well described by a polynomial of sufﬁciently high order. It

has been suggested recently that some of the functions appearing in fractional calculus and, in particular, some of the func-

tions emerging as solutions to fractional differential equations are better approximated using fractional power series

X

n

f ðtÞ ¼ ai t bi þ Rnþ1;b ðtÞ; ð3Þ

i¼0

known as the fractional Taylor expansion [28], which is also utilized in [29]. A related research is presented also [30,31]. It

seems reasonable, therefore, to seek numerical integration schemes which are particularly suitable for such functions. In the

sequel, a polynomial in tb , with b > 0,

X

n

Pn ðt b Þ ¼ Pn;b ðtÞ ¼ ai t bi ð4Þ

i¼0

is denoted as a quasi-polynomial of order n with commensurate power b. We will also use a shorter, but slightly less pre-

cise term b-polynomial of order n.

In this paper, the problem of numerical evaluation of fractional integral with unit upper integration limit

Z 1

1

a

0 I1 f ¼ f ðsÞð1 sÞa1 ds ð5Þ

CðaÞ 0

is considered. By means of simple scaling of the upper integration limit t, the general problem of evaluating (2) can easily be

reduced to the problem of evaluating (5). In particular, it can readily be shown that

a

0 It f ¼ ta 0 Ia1 f s ; f s ðsÞ ¼ f ðstÞ: ð6Þ

In order to evaluate (5), a family of orthogonal sets of quasi-polynomials with arbitrary commensurate powers

b > 0 is designed ﬁrst. The orthogonality is understood in the sense of fractional integration of order a with upper limit equal

to 1,

a ðaÞ ðaÞ

0 I1 Pn;b Pm;b ¼ 0; n – m; ð7Þ

208 M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219

ðaÞ

where P n;b denotes orthogonal b-polynomial of order n. The symbol ðaÞ in the upper right index simply denotes the order of

the fractional integration used to deﬁne orthogonality.

Second, using designed orthogonal quasi-polynomials, a family of Gaussian quadrature formula

a

X

n

ða Þ ðaÞ

0 I1 f Ai;b f ðti;b Þ ð8Þ

i¼0

is proposed. A formula of order n can be chosen from this family such that it is accurate for any b-polynomial integrand of

ðaÞ ðaÞ

order 2n þ 1 or less. Both quadrature coefﬁcients Ai;b and grid points t i;b are functions of the integration order a and the com-

mensurate power b. Thus, the commensurate power can be chosen in such a way that the integration error is minimized,

even if the considered class of integrands can not be represented exactly by a quasi-polynomial of a ﬁnite order.

The rest of the paper is organized as follows. Interpolation by means of quasi-polynomials is considered in the following

Section 2. Generalized Hermitean quadratures are considered and an error estimate is provided by Theorem 1. A short com-

ment on the quasi-polynomial Lagrange interpolation is also given. Hermitean and Gaussian quasi-polynomial quadratures

are developed in Section 3. The quadrature error has been investigated in Theorem 2. The main results of the paper are pre-

sented in Sections 4 and 5, presenting a new class of orthogonal polynomials and related novel Gaussian quadrature formu-

las, respectively. The existence of the proposed orthogonal quasi-polynomial families is established by Theorem 3, and the

localization of their roots by Theorem 4. Concluding remarks are given in Section 6.

Let us consider the problem of approximating a mapping f : ½0; T ! R with known function values f ðti Þ ¼ fi and values of

the ﬁrst derivatives dtd f ðt i Þ ¼ g i at distinct node points t i ; i 2 f0; . . . ; ng, such that

0 6 t 0 < < t n 6 T:

Consider the auxiliary function

pi;b ðtÞ ¼ ; ð9Þ

ðtbi t b0 Þ ðt bi t bi1 Þðt bi t biþ1 Þ ðt bi t bn Þ

where i 2 f0; . . . ; ng and b > 0. It is not difﬁcult to see that pi;b is a b-polynomial of order n, with zeros located at interpolation

nodes tj (j – i). More precisely,

1; i ¼ j;

pi;b ðt j Þ ¼ di;j ¼ ; ð10Þ

0; i–j

with di;j denoting the Kronecker d-symbol. Let us also deﬁne auxiliary functions

!

d

pi;b ðt i Þ t b t bi 2

ui;b ðtÞ ¼ 1 2 dt pi;b ðtÞ ; ð11Þ

b t b1

i

2

t b t bi pi;b ðt Þ

v i;b ðtÞ ¼ ð12Þ

b t b1

i

d

ui;b ðtj Þ ¼ di;j ; ui;b ðt j Þ ¼ 0; ð13Þ

dt

d

v i;b ðtj Þ ¼ 0; v i;b ðtj Þ ¼ di;j : ð14Þ

dt

The appropriate interpolating b-polynomial of order 2n 1 is

X

n X

n

Hn;b ðtÞ ¼ ui;b ðtÞfi þ v i;b ðtÞg i : ð15Þ

k¼0 k¼0

Quasi-polynomial (15) is denoted as the interpolating Hermite b-polynomial. An example of Hermitean interpolation

with b ¼ 0:5 is presented in Fig. 1. Note that the same set of points, with the same derivative values, could have been

interpolated using quasi-polynomial of any commensurate order b. The behavior of such approximations between the

node t i points would differ considerably. Different choices of b may be particularly suitable for each speciﬁc range of

applications.

M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219 209

Fig. 1. Hermite quasi-polynomial interpolating points ð1; 1Þ; ð2; 1Þ and ð3; 1Þ and with derivative values 1; 1 and 0, respectfully. The commensurate

order b ¼ 12 was utilized. The derivative of the interpolating polynomial is also shown.

f ðxÞ ¼ f ðp ﬃﬃﬃ

b

xÞ; for all x 2 ½0; T b ð16Þ

is n-times differentiable.

(b) The nth derivative of the auxiliary f is bounded with upper bound M, i.e.

n

dx f ðxÞ

< M: ð17Þ

Theorem 1. Assume that f : ½0; T ! R is of class B2nþ2;b ðMÞ. Assume that f has been evaluated, together with its ﬁrst derivative, at

n þ 1 distinct points t i , and let f ðti Þ ¼ fi and dtd f ðti Þ ¼ g i for all i 2 f0; . . . ; ng. The following statements hold for the Hermitean inter-

polating polynomial (15):

(a) Hn;b ðtÞ is the unique b-polynomial of order 2n þ 1 such that Hn;b ðti Þ ¼ f ðt i Þ ¼ fi and dtd Hn;b ðt i Þ ¼ f_ ðt i Þ ¼ g i for all

i 2 f0; . . . ; ng

(b) Let us denote the approximation error by EHn;b ðtÞ ¼ f ðtÞ Hn;b ðtÞ. The following error bound holds

M 2

jEHn;b ðtÞj 6 pn;b ðtb Þ ð18Þ

ð2n þ 2Þ!

with

pnþ1;b ðtÞ ¼ ðtb tb0 Þðtb tb1 Þ ðtb tbn Þ: ð19Þ

pﬃﬃﬃ

Proof. Introduce the auxiliary polynomial f ðxÞ ¼ f ð b xÞ and xi ¼ t bi . Note that f ðxi Þ ¼ f ðti Þ ¼ fi and also that

pﬃﬃﬃ

df ðxÞ df ð b xÞ df ðtÞ 1 1b1 df ðtÞ 1

¼ ¼ x ¼ : ð20Þ

dx dx dt b dt bt b1

Consequently

df ðxi Þ df ðt i Þ 1 1

¼ ¼ g i b1 : ð21Þ

dx dt bt b1

i bt i

It can readily be veriﬁed that by introducing the substitution x ¼ t b into the classical Hermitean polynomial interpolating

f ðxÞ, one obtains the Hermitean b-polynomial interpolating f ðtÞ. To see this, note that

210 M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219

dp ðxi Þ

i ðxÞ ¼

u 1 2ðx xi Þ i ðpi ðxÞÞ2 ; ð22Þ

dx

!

dpi;b ðti Þ 1

¼ 1 2ðt b t bi Þ ðpi;b ðtÞÞ2 ¼ ui;b ðtÞ ð23Þ

dt btb1

i

and similarly,

i v i;b ðtÞ: ð24Þ

df ðxi Þ

The classical interpolating polynomial of f , such that f ðxi Þ ¼ fi and dx

1

¼ g i btb1 is

i

X

n X

n

1

n ðxÞ ¼

H ui ðxÞfi þ v i ðxÞgi :

b1

i¼0 i¼0 bt i

n ðxÞ.

Introducing the appropriate substitutions, one readily obtains Hn;b ðtÞ ¼ H

The uniqueness of such an interpolating polynomial now follows as a classical result. The approximation error can be

found by noting that

H ðxÞ ¼ f ðtÞ H

EHn;b ðtÞ ¼ f ðtÞ Hn;b ðtÞ ¼ E n ðtÞ; ð25Þ

n

df

assuming naturally that x ¼ t ; xi ¼ b

tbi , and also that dx

1 b

¼ g i btb1 . By a classical result, there exists n 2 ½0; T such that

i

d 2nþ2

f ðnÞ

H ðxÞ ¼ dx

En ðpn ðxÞÞ2 :

ð2n þ 2Þ!

d 2nþ2

dx f ðnÞ

< M and pn ðxÞ ¼ pn;b ðtÞ. h

Remark 1. Assuming that the interpolating quasi-polynomial needs to be computed on the basis of function values only, it is

not hard to see that the appropriate interpolating quasi-polynomial is

X

n

Ln;b ðtÞ ¼ pi;b ðtÞfi : ð26Þ

i¼0

We refer to (26) as the Lagrange interpolating b-polynomial. By simple consideration, it is not hard to see that the approx-

imation error ELn;b ðtÞ ¼ f ðtÞ Ln;b ðtÞ is bounded by

M

jELn;b ðtÞj 6 jpnþ1;b ðtÞj ð27Þ

ðn þ 1Þ!

provided that f 2 Bnþ1;b ðMÞ, for some M > 0.

By means of a classical Hermitean formula, the value of an integral of a function is evaluated numerically by exact inte-

gration of the corresponding Hermitean interpolating polynomial. In the present paper, a generalization of this approach to

the case of fractional integration is presented. The proposed formulas are obtained by exact fractional integration of the cor-

responding Hermitean quasi-polynomial.

Consider a real-valued mapping f : ½0; T ! R. Assume that values of f and its ﬁrst derivative dtd f are known at distinct

points t i ; i 2 f0; . . . ; ng, and let f ðt i Þ ¼ fi and dtd f ðt i Þ ¼ g i . Quadrature formula

a

X

n

0 IT f Ai;b;a fi þ Bi;b;a g i ð28Þ

i¼0

will be called Hermitean quasi-polynomial quadrature if it gives exact values of the fractional integral of order a for f being

any b-polynomial of order less or equal to 2n 1.

By direct evaluation of the fractional integral of order a of the Hermitean approximation (15), one obtains

Ai;b;a ¼ 0 IaT ui;b ; ð29Þ

The following Theorem gives an error bound when using (28)–(30) to numerically compute fractional integrals of non-quasi-

polynomial integrands.

M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219 211

Theorem 2. Denote by E H n;b the error when using (28)–(30) to evaluate fractional integral of order

P

a

a; EH ¼ 0 IT f ni¼0 Ai;b;a fi þ Bi;b;a g i . If f : ½0; T ! R is a function of class B2nþ2;b ðMÞ then the following upper bound on the

n;b

approximation error holds

M 2

jE Hn;b j 6 j0 Ia pn;b ðt b Þ j: ð31Þ

ð2n þ 2Þ! T

Proof. The claim follows immediately by noting that the integration error is in fact equal to the fractional integral of the

approximation error (18). h

Gaussian quadratures are special case of Hermitean ones in which the quadrature nodes are selected so that the coefﬁ-

cients Bi;b;a multiplying derivative values of the integrand are equal to zero. Thus, by choosing quadrature nodes in a speciﬁc

manner, one obtains a quadrature formula of order 2n þ 1 using only integrand values at n þ 1 distinct nodes.

In particular, if

" 2 #

tb t bi pi;b ðt Þ

Bi;b;a ¼ 0 IaT v i;b ¼ 0 IaT ¼ 0; ð32Þ

b t b1

i

a

X

n

ðaÞ

0 IT f Ai;b fi ð33Þ

i¼0

ðaÞ

with quadrature coefﬁcients Ai;b ¼ Ai;b;a given by (29). Quadrature formula (33) will be denoted as Gaussian quasi-polyno-

mial quadrature formula.

The condition (32) implies that the quadrature nodes should be chosen as zeros of a special class of b-polynomials that are

orthogonal with respect to the a-integration operator. Such orthogonal polynomials are further investigated in the following

Section 4.

ða Þ

Remark 2. In a numerical setting, direct evaluation of quadrature coefﬁcients Ai;b using (29) is often not appropriate. It is

much more convenient to require that (33) hold for all b-monomials of order up to 2n 1,

a jb Cðjb þ 1Þ X

n

ðaÞ

0 I1 t ¼ ¼ A t jb ; j ¼ 0; . . . ; N 1: ð34Þ

Cðjb þ a þ 1Þ i¼0 i;b i

In the above expression, ti are the grid points. In an error-free setting expression 34 would be valid for N ¼ 2n, and one could

ðaÞ

choose arbitrary n equations to solve for Ai;b . In practice, due to round-off errors and similar computational imperfections,

Eq. (34) are satisﬁed with small errors. Thus, it is best to use all 2n þ 1 equation and solve them in the mean-square sense.

Quasi-polynomials orthogonal with respect to the fractional integration operator 0 Ia1 will be considered next. A procedure

for explicit calculation of quasi-polynomial coefﬁcients will be presented ﬁrst in the following Section 4.1. An existence

results will be presented in the subsequent Section 4.2, together with a convenient result claiming that all roots of such

quasi-polynomials are simple and located within ð0; 1Þ. Finally, some examples of orthogonal quasi-polynomials are pre-

sented in Section 4.3.

ðaÞ ðaÞ ðaÞ

Pn;b ðtÞ ¼ t nb þ an;b;n1 tðn1Þb þ þ an;b;0 ð35Þ

is orthogonal with respect to the fractional integration operator 0 Ia1 , i.e. the conditions under which (7) is satisﬁed. Without

loss of generality, all polynomials are assumed to be monic, i.e. the leading coefﬁcient is assumed to be one. Also without loss

of generality, only orthogonality with respect to fractional integration on the unit interval is considered. b-polynomials frac-

tionally orthogonal over more general intervals are obtained by simple scaling, as demonstrated by (6).

To emphasize the generality of our results, and also to enhance clarity of the presentation, the coefﬁcient of the orthog-

ðaÞ

onal b-polynomial of order n multiplying factor t ib is denoted as an;b;i . The upper right index ðaÞ denotes that the orthogonal-

ity is understood in the sense of the fractional integral of order a.

212 M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219

It is clear that a b-polynomial of order n will be orthogonal to all b-polynomials of lower order if and only if it is orthog-

onal to all b-monomials of order up to n 1. The condition (7) can thus be replaced by

a ðaÞ

0 I1 Pn;b ðtÞt ib ¼ 0; i < n: ð36Þ

ðaÞ

In order to obtain explicit expressions for evaluation of coefﬁcients an;b;i , let us introduce

Z 1

1

MðcaÞ ¼ 0 Ia1 t c ¼ tc ð1 tÞa1 dt: ð37Þ

CðaÞ 0

By recalling the deﬁnition and basic properties of Euler’s Beta function [8,32]

Z 1

CðxÞCðyÞ

Bðx; yÞ ¼ t x1 ð1 tÞy1 dt ¼ ; ð38Þ

0 Cðx þ yÞ

it readily follows that

1 Cðc þ 1Þ

MðcaÞ ¼ Bðc þ 1; aÞ ¼ : ð39Þ

CðaÞ Cðc þ a þ 1Þ

By utilization of (37), the condition (36) reduces to

ðaÞ ðaÞ ðaÞ ða Þ ðaÞ

an;b;0 M ib þ . . . þ an;b;n1 Mðiþn1Þb þ M ðiþnÞb ¼ 0; i < n: ð40Þ

ða Þ ðaÞ

Mn;b aan;b ¼ mn;b ; ð41Þ

where

2 ða Þ ðaÞ

3

M0 Mðn1Þb

6 7

ðaÞ 6 .. .. .. 7

Mn;b ¼6 . . . 7; ð42Þ

4 5

ðaÞ ða Þ

M ðn1Þb Mð2n2Þb

2 ðaÞ

3 2 ðaÞ

3

an;b;0 MðnÞb

6 7 6 7

6 .. 7 ðaÞ 6 .. 7

aan;b ¼6 . 7; mn;b ¼6 . 7: ð43Þ

4 5 4 5

ðaÞ ðaÞ

an;b;n1 M ð2n1Þb

ðaÞ

It is clear now that the desired family of orthogonal polynomials exists if and only if matrices Mn;b are regular (invertible). In

the present work, however, the existence of the orthogonal quasi-polynomials is investigated in an alternative way – by

reducing them to the classical orthogonal polynomials with suitably chosen weight function. This approach facilitate for

direct derivation of related recurrent polynomial relations and a number of associated properties. Expression (41) is still very

useful for numerical evaluation of b-polynomial coefﬁcients.

Denote by L2 ð0; 1Þ the set of square integrable real-valued functions in the Lebesgue sense over the interval ð0; 1Þ. Let

f 2 L2 ð0; 1Þ and g 2 L2 ð0; 1Þ be arbitrary, and let p 2 L2 ð0; 1Þ be non-negative (i.e. pðxÞ P 0 for all x).

The inner product of f and g with weight function p will be deﬁned as

Z 1

< f ; g>p ¼ pðxÞf ðxÞgðxÞdx: ð44Þ

0

A family of polynomials P k (k ¼ 0; 1; . . .) is said to be orthogonal in the sense of (44) if for all i – j hPi ; Pj ip ¼ 0.

The following classical constructive result establish the existence of family of classical polynomials orthogonal with

respect to the inner product (44).

Lemma 1 [33]. For any non-negative valued p 2 L2 ða; bÞ there exists a family of monic polynomials P k (k ¼ 0; 1; . . .) such that

hPi ; P j ip ¼ 0; i – j: ð45Þ

P0 ðxÞ ¼ 1; ð46Þ

M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219 213

!

hxPk ; Pk ip hPk ; Pk ip

Pkþ1 ðxÞ ¼ x Pk ðxÞ Pk1 ðxÞ: ð47Þ

hPk ; Pk ip hPk1 ; Pk1 ip

The existence of the orthogonal set of quasi-polynomials can now be readily established.

ða Þ

Theorem 3. There exists a family of b-polynomials Pn;b ðtÞ ¼ PnðaÞ ðtb Þ satisfying (7). These quasi-polynomials can be obtained

recursively by means of

ðaÞ

P0 ðxÞ ¼ 1; ð48Þ

!

ðaÞ hxPk ; Pk ip ðaÞ hPk ; Pk ip ðaÞ

Pkþ1 ðxÞ ¼ x Pk ðxÞ P ðxÞ ð49Þ

hPk ; Pk ip hPk1 ; Pk1 ip k1

with

1 a1

1 ð1 xb Þ

pðxÞ ¼ : ð50Þ

CðaÞ b1

bx b

Z 1

1

PðnaÞ ðt b ÞPðmaÞ ðt b Þð1 sÞa1 ds ¼ 0; m – n:

CðaÞ 0

Z 1

1 1 a1 dx

PðnaÞ ðxÞPðmaÞ ðxÞð1 xb Þ ¼ 0; m – n:

CðaÞ 0 bx

b1

b

It is now obvious that orthogonality condition (7) reduces to the classical orthogonality condition (45), with weight function

ðaÞ

p given by (50). Note also that this weight is square-integrable on (0, 1) and non-negative, thus P k ðxÞ exists as classical

orthogonal polynomials according to Lemma 1. h

Remark 3. Note that the recursive relation (49) can be rewritten in a more compact form, without explicit occurrence of the

weight function p given by (50). In fact, since

ðaÞ ðaÞ

hP ðmaÞ ; PðnaÞ ip ¼ 0 Ia1 Pm;b Pn;b

0 1

ðaÞ ðaÞ

a a ðaÞ ðaÞ

t b Pk;b Pk;b

0 I1 0 I 1 P k;b P k;b

ðaÞ a aÞ

¼ @t APk;b ðtÞ

ð Þ

Pkþ1;b ðtÞ b

P ðk1 ðxÞ:

a ðaÞ ðaÞ a ðaÞ ðaÞ

I

0 1 P P

k;b k;b I

0 1 P k 1; bP k1;b

Lemma 2 [33]. All roots of monic polynomial Q k belonging to orthogonal family with respect to (45) are real, simple, and lie in

ð0; 1Þ.

As a direct consequence, one can easily prove the following claim.

ða Þ

Theorem 4. Each orthogonal quasi-polynomials Pk ðtb Þ has exactly k real roots, each of which is simple and lie in ð0; 1Þ.

ðaÞ

Proof. Since Pk ðxÞ is a classical orthogonal polynomial, Lemma 2 claims that all of its roots are real, simple and lie in ð0; 1Þ.

The claim now follows by noting that x ¼ tb , which is a bijective function mapping ð0; 1Þ to itself. h

For illustration purposes, let us explicitly compute ﬁrst few orthogonal quasi-polynomials for some speciﬁc values of inte-

gration order a, and commensurate polynomial order b. Classical polynomials (b ¼ 1) orthogonal with respect to the frac-

tional integration of order a are given in (51).

214 M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219

ðaÞ

P0;1 ðtÞ ¼ 1;

ðaÞ 1

P1;1 ðtÞ ¼ þ t;

aþ1

ðaÞ 2 4

P2;1 ðtÞ ¼ t þ t2 ; ð51Þ

ða þ 2Þða þ 3Þ a þ 3

ðaÞ 6 18 9 2

P3;1 ðtÞ ¼ þ t t þ t3 ;

ða þ 3Þða þ 4Þða þ 5Þ ða þ 4Þða þ 5Þ aþ5

..

.

Quasi-polynomials with different commensurate order b orthogonal with respect to the classical integral (a ¼ 1) are pre-

sented in (52).

ð1Þ

P0;b ðtÞ ¼ 1;

ð1Þ 1

P1;b ðtÞ ¼ þ tb ;

bþ1

P2;b ðtÞ ¼ 2

t þ t 2b ; ð52Þ

6b þ 5b þ 1 3b þ 1

P3;b ðtÞ ¼ þ t t þ t3b ;

3

60b þ 47b þ 12b þ 1 2 ð4a þ 1Þð5a þ 1Þ 5b þ 1

..

.

Quasi-polynomials of order 1, 2, and 3 with commensurate order b ¼ 12, orthogonal with respect to semi-integral (a ¼ 12) are

presented in Fig. 2.

Considerations of the last two sections enable us to construct a continuum of approximate fractional integration formu-

las. First, by means of (6), each numerical integration problem is ﬁrst reduced to the equivalent problem in which the upper

terminal is 1. Second, given any positive real a, a quadrature formula

a

X

n

ða Þ ðaÞ

0 I1 f Ai;b f ðti;b Þ ð53Þ

i¼0

Fig. 2. (Left) Quasi-polynomials of commensurate order b ¼ 12, orthogonal with respect to the fractional integral of order a ¼ 12 over interval ð0; 1Þ. (Right)

Quasi-polynomials of commensurate order b ¼ 14, orthogonal with respect to the classical ﬁrst order integral (a ¼ 1) over interval the same interval.

Diagrams show quasi-polynomials with different order n. Notice that the quasi-polynomial of order n has exactly n zero-crossings on ð0; 1Þ.

M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219 215

1. Choose b such that the integrand under consideration is naturally approximated by a b-polynomial. Choose n so that

the integration error given by Theorem 2 is acceptable.

ðaÞ

2. Compute the coefﬁcients of a b-polynomial P n;b by solving the system of linear equations given by (41), or equiva-

lently by solving recurrence (48) and (49).

ðaÞ ðaÞ

3. Compute the quadrature nodes ti;b (i 2 f0; 1; . . . ; ng) as roots of the quasi-polynomial P n;b computed in the previous

step.

ðaÞ

4. Compute quadrature coefﬁcients Ai;b (i 2 f0; 1; . . . ; ng) according to (29). Equivalently, one can also obtain quadrature

coefﬁcients by solving system of linear equation (34). The later is the preferred approach in a numerical setting.

5. Apply (53) to obtain the desired approximation.

ðaÞ

Step 2 of the proposed scheme, the evaluation of the appropriate b-polynomial P n;b , can be computed in a numerically

ðaÞ

efﬁcient and stable manner. The same is also true for the computation of the quadrature coefﬁcients Ai;b in Step 4. Step 3,

however, involves ﬁnding zeros of b-polynomials. As such, it is numerically unstable and must be approached with care.

One solution would be to embed our approach into the classical theory of Gaussian integration. Notably, by substitution

x ¼ tb and introducing the weight function (50) one is free to utilize the Golub–Welsch algorithm [34] which provides a

numerically stable and well-conditioned alternative. More recent related work can for example be found in publications

of Milovanović and co-authors [35–38,3] and others, i.e. [39,40]. The related software for precise evaluation of Gaussian

quadratures is described in [41,42], (see also www.mi.sanu.ac.rs/gvm/orthogonalpolynomials.php).

ðaÞ

In a numerical setting, high degrees of accuracy can be obtained by storing the desired grid-points ti;b and quadrature

ðaÞ

coefﬁcients Ai;b in advance using some high-precision computational engine. The quadrature formula in Step 4 can then

be evaluated safely and efﬁciently even on a standard, double precision numerical platform. The examples in the sequel will

reveal, however, that by proper choice of the commensurate order b good accuracy can be obtained even in a pure IEEE dou-

ble-precision numerical environment (i.e. by implementing all steps of the proposed scheme without variable-precision

numerical engine such are MAPLE or MATHEMATICA).

Consider a function

pﬃﬃ pﬃﬃ

f ðtÞ ¼ et erfc t ¼ E1;1 t ð54Þ

2

with erfc and Ea;b being the well-known complementary error-function and two-parameter Mittag–Lefﬂer function, respec-

tively [6]. Functions of this form are quite common in the ﬁeld of fractional calculus. In particular, (54) is the step-response of

a linear process described by a fractional transfer function

Fig. 3. Logarithm of the absolute integration error when evaluating semi-integral (a ¼ 12) function (54) with quadrature formula (53) using different choices

of b. Figure in the left shows accuracy of the quadrature formula of order n ¼ 1, where the ﬁgure in the right depict accuracy of the quadrature formula of

order n ¼ 3.

216 M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219

pﬃﬃ

s

GðsÞ ¼ pﬃﬃ : ð55Þ

sþ1

The semi-integral of (54) can readily be evaluated analytically,

1

2

X

1

tk

Ea;b ðtÞ ¼ ; ð57Þ

k¼0

Cðak þ bÞ

pﬃﬃ X1

pﬃﬃk

t

f ðtÞ ¼ E1;1 t ¼ : ð58Þ

2

k¼0

C 2k þ 1

Fig. 4. Logarithm of the absolute integration error when evaluating ﬁrst-order integral (a ¼ 1) function (54) with quadrature formula (53) using different

choices of b. Figure in the left shows accuracy of the quadrature formula of order n ¼ 1, where the ﬁgure in the right depict accuracy of the quadrature

formula of order n ¼ 3.

Fig. 5. Logarithms of absolute errors for various quadrature orders and for different values of b when evaluating semi-integral of (54). All approximations

were obtained using standard, IEEE double precision arithmetic.

M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219 217

Table 1

Logarithm of the absolute error when evaluating 0 I0:5

1 J 0:5 ðtÞ, where J n ðtÞ is the Bessel function of the ﬁrst kind with. All computations were performed using IEEE

double precision arithmetic.

b ¼ 0:1 2.589 2.081 2.853 3.884 5.443 5.401 6.072 6.974 6.578 8.163

b ¼ 0:2 2.008 2.261 3.418 5.113 5.485 6.636 8.659 9.021 10.116 9.987

b ¼ 0:3 1.791 2.486 4.286 5.086 6.423 8.040 8.526 8.951 9.390 9.757

b ¼ 0:4 1.662 2.748 4.460 5.392 6.285 6.794 7.231 7.617 7.962 8.288

b ¼ 0:5 1.572 3.001 4.634 6.661 8.252 10.858 12.220 15.477 15.955 15.477

b ¼ 0:6 1.504 3.091 4.220 4.697 5.131 5.491 5.798 6.067 6.306 6.517

b ¼ 0:7 1.451 2.951 3.643 4.119 4.498 4.812 5.080 5.314 5.522 5.700

b ¼ 0:8 1.408 2.739 3.304 3.733 4.070 4.349 4.587 4.794 4.978 5.148

b ¼ 0:9 1.372 2.547 3.057 3.443 3.747 3.998 4.212 4.398 4.562 4.705

b ¼ 1:0 1.341 2.387 2.861 3.214 3.491 3.719 3.913 4.082 4.232 4.388

Table 2

Logarithm of the absolute error when evaluating 0 Ia1 J0:5 ðtÞ with ﬁxed b ¼ 0:25. Jn ðtÞ is the Bessel function of the ﬁrst kind with. All computations were performed

using IEEE double precision arithmetic.

a ¼ 0:1 1.980 2.831 4.376 5.395 6.459 8.244 9.197 10.369 12.036 12.990

a ¼ 0:2 1.825 2.585 4.090 5.152 6.194 7.948 8.942 10.103 11.895 12.771

a ¼ 0:3 1.796 2.467 3.935 5.038 6.059 7.784 8.817 9.964 11.769 12.540

a ¼ 0:4 1.822 2.402 3.837 4.979 5.980 7.678 8.747 9.879 11.619 12.585

a ¼ 0:5 1.883 2.368 3.773 4.951 5.931 7.605 8.709 9.828 11.586 12.728

a ¼ 0:6 1.973 2.353 3.731 4.945 5.904 7.556 8.693 9.797 11.512 12.576

a ¼ 0:7 2.092 2.351 3.705 4.953 5.892 7.523 8.691 9.782 11.490 12.690

a ¼ 0:8 2.250 2.360 3.691 4.973 5.891 7.503 8.702 9.779 11.487 12.860

a ¼ 0:9 2.474 2.377 3.687 5.003 5.900 7.494 8.722 9.786 11.474 12.707

a ¼ 1:0 2.863 2.401 3.692 5.040 5.916 7.493 8.751 9.800 11.478 12.906

It seems natural to approximate this function using quasi-polynomials of orthogonal order 12. Fig. 3-Left shows the logarithm

of the absolute integration error when evaluating the semi-integral f using quasi-polynomials of order n ¼ 1 for various val-

ues of commensurate order b. As it is seen from the ﬁgure, the choice b ¼ 12 is indeed a solid one, yet not the best one avail-

11

able. A distinct error minimum is obtained for b 100 . A similar case is seen in Fig. 3-Right depicting the integration error in

the case n ¼ 3. There are multiple local error minimums, one of which actually being b ¼ 12. The best approximation is

15

obtained for b 100 . In both cases, a signiﬁcant error reduction can be obtained by the right choice of b. Notice also that

the classical choice of b ¼ 1 is, in fact, the worst one in this particular case. For n ¼ 3 the optimal choice of the commensurate

order reduces the integration error by approximately 6 orders of magnitude!

The proper choice of the commensurate order b is equally important when evaluating the classical, ﬁrst-order integrals

(a ¼ 1). This can clearly be seen from the diagrams in Fig. 4. Again, an optimal choice of b improves accuracy for several

orders of magnitude, and again the classical choice of b ¼ 1 is, in fact, the least favorable one.

In order to illustrate the effectiveness of the proposed schemes when utilized, logarithms of absolute errors for various

quadrature orders and for different values of b are shown in Fig. 5. The results shown in this ﬁgure were obtained using stan-

dard IEEE double precision arithmetic with machine accuracy 2:22 1016 .

To further illustrate the performance of the proposed scheme when applied using the standard IEEE double precision

arithmetic, consider Bessel function of the ﬁrst kind, J n . Accuracy of the proposed method when evaluating semi-integrals

of J n with quadrature orders up to 10, and for various values of b is shown in Table 1. For most values of b, error saturates

for large N. This is a direct consequence of accumulated round-off errors (note the oscillation of error values for b ¼ 0:5 and

N ¼ 8; 9 and 10). By choosing b properly errors are converging to machine accuracy. Note that the reported error levels are

quite acceptable for most engineering applications. If higher levels of accuracy are required, one must resort to high precision

arithmetic, and to techniques outlined in Section 5.2. Table 2 reports accuracy of the proposed method applied to J 0:5 ðtÞ for

ﬁxed value of b ¼ 0:25, but for various order of the integration order a.

6. Conclusions

A novel class of fractionally orthogonal quasi-polynomials has been developed and investigated. These quasi-polynomials

are then utilized for development of suitable Gaussian quadratures for numerical evaluation of fractional integrals. The pro-

posed quadratures are highly efﬁcient and ﬂexible, since the commensurate order of the interpolating quasi-polynomials can

be selected freely and ﬁne-tuned for a particular class of problems. As a secondary result, novel class of Lagrangian and

Hermitean interpolation formulas, as well as a class of Hermitean numerical integration formulas have been developed.

218 M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219

By choosing a ¼ 1 and selecting arbitrary values of the commensurate order b one can obtain novel formulas for the

numerical evaluation of integer order integrals as well. By an appropriate selection of the commensurate order one can

reduce the integration error considerably even in this classical case, as was demonstrated by an example of Section 5.2.

Consequent research will concentrate on the investigation of further properties of the developed class of fractionally

orthogonal quasi-polynomials, in particular further research will address their generating functions. Applications of the

Golub–Welsch algorithm for efﬁcient computation of the proposed quadrature formulas should also be investigated further.

It is also interesting to consider quasi-polynomials orthogonal to more general inﬁnite dimensional operators, as well as the

utilization of such polynomial in solving fractional differential equations, simulation and identiﬁcation of fractional order

systems.

Acknowledgments

The authors wish to express their gratitude to prof. dr Gradimir V. Milovanović, Member of the Serbian Academy of

Science and Arts, on his numerous suggestions and insights that considerably contributed to our work.

M.R. Rapaić and T.B. Šekara were partially supported by the Ministry of Education, Science and Technological Develop-

ment of Republic of Serbia, Grants Nos. TR32018 and TR33013. (MRR) and TR33020 (TBŠ).

References

[2] R.L. Burden, J.D. Faires, Numerical Analysis Textbook and Study Guide, seventh ed., Brooks & Cole, 2001.

[3] G. Mastroianni, G.V. Milovanović, Interpolation Processes – Basic Theory and Applications, Springer, 2008.

[4] J. Sabatier, O.P. Agrawal, J.A.T. Machado, Advances in Fractional Calculus, Springer, Berlin, 2007.

[5] B.J. West, M. Bologna, P. Grigolini, Physics of Fractal Operators, Springer, 2003.

[6] R.L. Magin, Fractional Calculus in Bioengineering, Begell-House, 2003.

[7] J.A.T. Machado, M.F. Silva, R.S. Barbosa, I.S. Jesus, C.M. Reis, M.G. Marcos, A.F. Galhano, Some applications of fractional calculus in engineering, Math.

Probl. Eng. 2010 (2010) 34, http://dx.doi.org/10.1155/2010/639801. Article ID 639801.

[8] I. Podlubny, Fractional Differential Equations, Academic Press, New York, 1999.

[9] K.B. Oldham, J. Spanier, The Fractional Calculus – Theory and Application of Differentiation and Interaction to an Arbitrary Order, Academic Press, San

Diego, 1974.

[10] K.S. Miller, B. Ross, An Introduction to Fractional Calculus and Fractional Differential Equations, John Willey and Sons, New York, 1993.

[11] A.I. Saichev, W.A. Woyczinsky, Distributions in Physical and Engineering Sciences, Birkhauser, 1997.

[12] K. Diethelm, N.J. Ford, A.D. Freed, Y. Luchko, Algorithms for the fractional calculus: a selection of numerical methods, Comput. Methods Appl. Mech.

Eng. 194 (2005) 743–773.

[13] Z. Odibat, Approximations of fractional integrals and Caputo fractional derivatives, Appl. Math. Comput. 178 (2006) 527–533.

[14] M.R. Rapaić, A. Pisano, Z.D. Jeličić, Trapezoidal rule for numerical evaluation of fractional order integrals with applications to simulation and

identiﬁcation of fractional order systems, in: Proceedings of IEEE MSC’12.

[15] A. Pisano, R. Garrapa, D. Nessi, M.R. Rapaić, Pseudo-recursive trapezoidal rule for the numerical solution of linear fractional differential equations, in:

Proceedings of FDA’13.

[16] U. Lepik, Solving fractional integral equations by the haar wavelet method, Appl. Math. Comput. 214 (2009) 468–478.

[17] Z. Gao, X. Liao, Discretization algorithm for fractional order integral by haar wavelet approximation, Appl. Math. Comput. 218 (2011) 1917–1926.

[18] M. Weilbeer, Efﬁcient numerical methods for fractional differential equations and their analytical background (Ph.D. thesis), Von der Carl-Friedrich-

Gauß-Fakultät für Mathematik und Informatik der Technischen Universität Braunschweig, 2005.

[19] K. Diethelm, J.M. Ford, N.J. Ford, M. Weilbeer, Pitfalls in fast numerical solvers for fractional differential equations, J. Comput. Appl. Math. 186 (2006)

482–503.

[20] T. Hasegawa, H. Sugiura, Uniform approximation to fractional derivatives of functions of algebraic singularity, J. Comput. Appl. Math. 228 (2009) 247–

253.

[21] H. Sugiura, T. Hasegawa, Quadrature rule for abels equations: uniformly approximating fractional derivatives, J. Comput. Appl. Math. 223 (2009) 459–

468.

[22] I. Podlubny, I. Petráš, B.M. Vinagre, P. O’Leary, L. Dorčák, Analogue realizations of fractional-order controllers, Nonlinear Dyn. 29 (2002) 281–296.

[23] B.M. Vinagre, Y.Q. Chen, I. Petráš, Two direct tustin discretization methods for fractional-order differentiator and integrator, J. Franklin Inst. 340 (2003)

349–362.

[24] R.S. Barbosa, J. Tenreiro-MAchado, Implementation of discrete-time fractional-order controllers based on LS approximations, Acta Polytech. Hungarica

3 (4) (2006) 5–22.

[25] M.R. Rapaić, T.B. Šekara, Novel direct optimal and indirect method for discretization of fractional systems, Electr. Eng. 93 (2011) 91–102.

[26] M. Aoun, R. Malti, F. Levron, A. Oustaloup, Synthesis of fractional Laguerre basis for system approximation, Automatica 43 (2007) 1640–1648.

[27] H. Akçay, Synthesis of complete orthonormal fractional basis functions with prescribed poles, IEEE Trans. Signal Process. 56 (2008) 4716–4728.

[28] D. Usero, Fractional Taylor series for Caputo fractional derivatives. Construction of numerical schemes.

[29] O.P. Agrawal, A numerical scheme and error analysis for a class of fractional optimal control problems, in: Proceedings of the ASME 2009 International

Design Engineering Technical Conferences & Computers and Information in Engineering Conference.

[30] M.-F. Li, J.-R. Ren, T. Zhu, Series expansion in fractional calculus and fractional differential equations. Available from: <arXiv:0910.4819>.

[31] J. Munkhammar, Riemann–Liouville fractional derivatives and the Taylor–Riemann series, Department of Mathematics, Upsala University Tech. rep.,

2004.

[32] A.A. Kilbas, H.M. Srivastava, J.J. Trujillo, Theory and Applications of Fractional Differential Equations, Elsevier, 2006.

[33] D. Radunović, Numeričke Metode, Akademska misao, Beograd, 2004.

[34] G.H. Golub, J.H. Welsch, Calculation of Gauss quadrature rules, Math. Comput. 23 (106) (1969) 221–230.

[35] G.V. Milovanović, Quadrature processes and new applications, Bull. Cl. Sci. Math. Nat. Sci. Math. 38 (38) (2013) 83–120.

[36] G.V. Milovanović, Müntz orthogonal polynomials and their numerical evaluation, in: W. Gautschi, G. Golub, G. Opfer (Eds.), Applications and

Computation of Orthogonal Polynomials, vol. 131, Birkhäuser, Basel, 1999, pp. 179–202.

[37] G. Milovanović, A. Cvetković, Nonstandard gaussian quadrature formulae based on operator values, Adv. Comput. Math. 32 (2010) 431–486.

[38] G. Milovanović, A. Cvetković, Gaussian type quadrature rules for Müntz systems, SIAM J. Sci. Comput. 27 (2005) 893–913.

[39] E. Shahrokh, M. Shamsi, Y. Luchko, Numerical solution of fractional differential equations with a collocation method based on müntz polynomials,

Comput. Math. Appl. 62 (3) (2011) 918–929.

M.R. Rapaić et al. / Applied Mathematics and Computation 245 (2014) 206–219 219

[40] S. Esmaeili, M. Shamsi, A pseudo-spectral scheme for the approximate solution of a family of fractional differential equations, Commun. Nonlinear Sci.

Numer. Simul. 16 (9) (2011) 3646–3654.

[41] A. Cvetković, G. Milovanović, The mathematica package orthogonal polynomials, Facta Univ. Ser. Math. Inf. 19 (2004) 17–36.

[42] G. Milovanović, A. Cvetković, Special classes of orthogonal polynomials and corresponding quadratures of Gaussian type, Math. Balkanica 26 (2012)

169–184.

- PP & QM IntegrationUploaded byPuneet Puri
- Article Methode Wilson TethaUploaded byEntps Dib
- dfht kim mmmUploaded bylaki pejic
- Numerical integration 1d elementUploaded byAmit Kaushal
- area under a curve - rectangle format 12-9-2017Uploaded byapi-323045358
- Final Assignment(Part 1)Uploaded bySukalyan Som
- sem1Uploaded bypraveennagarajan
- NumericalMethodsT264UnitIVByDrNVNagendram.pdfUploaded byAkhilesh kumar
- numerical technique for guideUploaded bymanny ansari
- Lecture 3 4Uploaded bySergi C. Cortada
- ISME 2008-A Truly Meshless Element Free Galerkin Method by a General Approach for Evaluation of Domain IntegralsUploaded byKevinPatrónHernandez
- abs_integrate.pdfUploaded byRikárdo Campos
- fem gaussUploaded byLana Todorović
- Theory of the Monte Carlo Method for SemiconductorUploaded bychoroampon
- Meshless ComputationUploaded bybardock277
- 1-2Uploaded bySiva Lrn Sivalrn
- Métodos NuméricosUploaded byAlam
- 1-IN-GATE18_schedule.pdfUploaded byAbhishek Agarwal
- PFUploaded bychinnasaib
- UT Dallas Syllabus for math2414.001.10f taught by Bentley Garrett (btg032000)Uploaded byUT Dallas Provost's Technology Group
- UT Dallas Syllabus for math2414.003.10f taught by Paul Stanford (phs031000)Uploaded byUT Dallas Provost's Technology Group
- Published - Carving TechniqueUploaded byKuldeep Bhattacharjee
- Pauls Online Math NotesUploaded bytan_siqi
- Old Testament Introduction and Sur VeyUploaded bypaul macharia
- math5Uploaded byArjun
- Mean value theoremUploaded byakirank1
- numerical methods finalsUploaded byHarris Anches
- 340222__.pdfUploaded byhadush
- CHE306Uploaded byAbiola Olawale
- Maple DuhamelUploaded byMatteo Soru

- Analiza bonitetaUploaded byDanijel Rodic
- Analiza bonitetaUploaded byDanijel Rodic
- Dynamic Scattering Function for High-temperature Liquid LeadUploaded byMatija Korvin
- Vrućinić Et Al. 2015 Advanced ScienceUploaded byMatija Korvin
- UsesandGratificationsinthe21stCentury.pdfUploaded byBerciu
- An Approximation of Phosphorescence Decay Kinetics of IdealUploaded byMatija Korvin
- Celinac_Zmajevac kompl SR (1).pdfUploaded byMatija Korvin
- Das Lasèguesche ZeichenUploaded byMatija Korvin
- A Novel Class of Fractionally Orthogonal Quasipolynomials and New Fractional Quadrature FormulasUploaded byMatija Korvin
- A Novel Class of Fractionally Orthogonal Quasipolynomials and New Fractional Quadrature FormulasUploaded byMatija Korvin
- A Novel Class of Fractionally Orthogonal Quasipolynomials and New Fractional Quadrature FormulasUploaded byMatija Korvin
- Mogućnosti iskorišćavanja mlijeka u brtsko-planinskim područjimaUploaded byMatija Korvin
- Celinac ZmajevacUploaded byMatija Korvin
- Effects of Different Maize Hybrids on Above Ground Biomass in Intercrops With SoybeanUploaded byMatija Korvin
- Primena Proaktivne Digitalne Forenzike u Cloud Computing OkruzenjuUploaded byMatija Korvin
- Использование Двуязычных Словарей и Комментариев Преподавателей в Переводных Заданиях в Целях Развития Словарного Запаса Английского Как Второго ЯзыкаUploaded byMatija Korvin
- PRIMENA NETEROVE TEOREME U MIKROPOLARNOJ MEHANICI KONTINUUMAUploaded byMatija Korvin
- Quality of life predictors in patients with chronic infammatory.pdfUploaded byMatija Korvin
- Quality of life predictors in patients with chronic infammatory.pdfUploaded byMatija Korvin
- Quality of life predictors in patients with chronic infammatory.pdfUploaded byMatija Korvin
- Razgovor o Ljevici 2015Uploaded byMatija Korvin
- An Approximation of Phosphorescence Decay Kinetics of IdealUploaded byMatija Korvin
- Very Fast Electromagnetic Transients in AirUploaded byMatija Korvin
- Tito and AstronautsUploaded byMatija Korvin
- Popis 1910 Po OpstinamaUploaded byMatija Korvin
- Veletic 2Uploaded byMatija Korvin
- Uvodna_studija_Markus.pdfUploaded byMatija Korvin
- Uvodna_studija_Markus.pdfUploaded byMatija Korvin
- Effect of Microwave Power and C2 Emission Intensity on Structural and SurfaceUploaded byMatija Korvin

- 12 Teresa Ponce de Leão LNEG PortugalUploaded byFrancisco Amoes
- Group 3 - Property Purchase StrategyUploaded bychelsea1989
- Cs Literature 120317101528 Phpapp01Uploaded bykothariankit03
- FM Statistics Holiday WorkUploaded byPawat Silawattakun
- PIANC Presentation on Approach channelUploaded byIgnatius Samraj
- Dimensional StabilityUploaded bySat ez
- Different Types of EssayUploaded byBambi Pajaro
- Bears, Bumblebees and Spiders in the Role of an ExpatriateUploaded byMianKhurramJohri
- Determination of Individual Competencies by Statistical Methods Yuksek Lisans Tezi Tulay Bozkurt 192215582006001Uploaded byTulay Bozkurt
- Descriptive Research - Research MethodologyUploaded byrajavendran
- Sobre ppseudo paneles.pdfUploaded byEdgar Cruz
- STI_2062Uploaded byfyzmfa
- AnovaUploaded byArun Beniwal
- EFAUploaded bySiti Khotimah
- Quick-R: Data TypesUploaded byChungil Chae
- Cialdini PrinciplesUploaded byOanabuzz
- Alabama School System Split AnalysisUploaded byTrisha Powell Crain
- Flash Eurobarometer 382a Europeans' Satisfaction With Rail ServicesUploaded bycavris
- 2006-CapeCoral-DataManagement-CaseUploaded byamit_vatsa
- Revision ExercisesUploaded byনীল জোছনা
- Interesting Standards ListUploaded byBoboskins Bee
- w16021.pdfUploaded byniloykrittika
- 6 Quality and Statistical Process Control Chapter 9Uploaded byTerry Conrad King
- f9Uploaded byjustlily
- Qualitative Research designUploaded bykimemia
- 013.pdfUploaded bySynths
- Liquid Limit Lab ReportUploaded byPoovan Rajaratnam
- Chapter -5 a Case Study on Environmental AuditingUploaded byJoel Sabastin Dass
- 10.1.1.794.7318.pdfUploaded byHafizAce
- iahs_201_0105Uploaded byramaraokolli4694