International Conference on Computer Aided Engineering (CAE-2013)

Department of Mechanical Engineering, IIT Madras, India

A NEW CLASS OF RATIONAL FRACTAL FUNCTION FOR CURVE FITTING

1

A. K. B. Chand1 , S. K. Katiyar1 and G. Saravana Kumar2

Department of Mathematics,2 Department of Engineering Design
Indian Institute of Technology Madras
Chennai, India - 600036

ABSTRACT
Fractal is a modern tool to model non-linear scientific and natural data effectively. Recent studies have developed the fractal
interpolation function (FIF) for modeling experimental and/or geometric data. The iterated function system (IFS) is used to
generate a FIF. The work deals with determination of IFS for approximating prescribed data using FIF. When the scaling
factors are restricted appropriately in an IFS, the resulting FIF is differentiable in nature. Shape preservation is significant
aspect for geometric modeling of curves and surfaces from data points. The shape preserving capabilities of polynomial
spline FIF are heavily limited. In the present paper, we introduce a new class of rational cubic FIF with shape preserving
properties. The developed rational IFS model uses cubic by cubic form with four shape parameters in each subinterval. This
scheme offers an additional freedom over the classical rational cubic interpolants due to the presence of scaling factors and
shape parameters. The classical rational cubic functions are obtained as a special case of the developed fractal interpolants.
The sufficiency conditions on the scaling factors and shape parameters are investigated and developed so that the resulting
rational cubic FIF is positive when the data set is positive. This approach offers a single specification for a large class of
positive fractal interpolants. The developed rational cubic spline FIF can be used for curve fitting if the data shows the trend
of a positive C 1 - smooth function, whereas its derivative is irregular. The study also shows that for appropriate values of the
scaling factors and shape parameters, the developed rational FIF converges uniformly to the data points coming from a C 3
continuous curve at least as rapidly as the third power of the mesh norm. A visual illustration of the shape preserving fractal
curves is provided to support our theoretical results.
Keywords: Fractal Interpolation Function, Iterated Function System, Rational Cubic Fractal functions, Shape Preservation,
Positivity, Convergence.

1. INTRODUCTION
Smooth curve representation of scientific data through
interpolation curve/surface has great significance in Computer
Aided Design, Computer Graphics, Information Sciences,
Geometric Modeling, Data visualization etc. However
applications of interpolation problem in CAGD, Cartography,
Image Analysis, etc., demand the interpolant to mimic
some geometric properties hidden in the data. The problem
of searching a sufficiently smooth function that preserves
the qualitative shape properties inherent in the data is
generally referred as shape preserving interpolation. The
shape properties are mathematically expressed in terms
of conditions such as positivity, monotony, and convexity.
Positivity is the most common and fundamental shape aspect.
For instance, positive data arise in resistance offered by an
electric circuit, progress of an irreversible process, monthly
rainfall amounts, level of gas discharge in certain chemical
reactions, volume, density etc. The negative graphical
representations of these physical quantities have no meaning.
The literature on classical shape preserving interpolation

chand@iitm.ac.in

where interpolants are polynomial, exponential and rational

spline is abundant,(see for instances [6], [8], [10], [12], [13]).
Similar to shape preservation, shape modification wherein
shape interpolant can be altered for a prescribed data is also
a requirement in design environment. Rational splines with
shape parameters are well suited for shape modification and
shape preservation. Rational splines have vast applications
in Reverse Engineering and Vectorization of object (CAD)
[9]. These traditional shape preserving interpolants with C r continuity is in fact infinitely differentiable except possibly
for finite number of data sets in the interpolation interval.
Therefore these methods are less satisfactory for preserving
the shape present in the data, where variables representing
derivatives are to be visualized by using functions having
irregularity in a dense subset of the interpolation interval,
for instance, the velocity of a particle undergoing Brownian
movement.
It is not ideal to use (piecewise) smooth interpolant when the
data has very irregular structure. Fractal interpolation function
(FIF) captures the irregularity of data very effectively in
comparison with the classical interpolants/splines. The FIF
provides an effective tool for modeling data sampled from

real world signals, graphs which are usually difficult to model

by using classical functions. Fractal interpolation offers
the possibility of choosing either smooth or non-smooth
approximants, as per problem requirement. Barnsley [1]
introduced the definition of Fractal Interpolation Function
(FIF) based on the theory of Iterated function system (IFS),
that consist of a set of mappings whose fixed point is the
graph of a continuous function that interpolates the data.
FIFs are used to approximate naturally occurring functions
which show some short of self similarity under magnification.
By imposing appropriate conditions on the scaling factors,
Barnsley et al. [2] observed that if the problem is of
differentiable type, then the elements of the IFS may be
suitably chosen so that corresponding FIF is smooth. But it
is difficult to get all types of boundary conditions for fractal
splines in this iterative construction. Fractal splines with
general boundary conditions have been studied in [3], [4] in
simpler ways.
We exploit the aforementioned flexibility and versatility of
fractal splines to overcome the stated limitations of existing
shape preserving rational splines. To be precise we shall
construct an IFS with component functions being rational
(x)
; where pi (x) and qi (x)
functions of the form si (x) = pqii (x)
are cubic polynomials. pi (x) are to be determined through
the Hermite interpolation conditions of the corresponding
FIF and qi (x) are preassigned cubic polynomials with four
shape parameters. It is observed that when the scaling factor
tends to zero and shape parameters tends to , the FIF
approaches piecewise linear interpolant. This assures that
FIF preserves shape of data in the limiting configuration.
Hence the problem is to find the extent to which scaling
factors can be deviated from zero and shape parameters
can be reduced from theoretical limit value so that the
corresponding interpolant preserves the shape of given data.
In this regard we have derived a set of sufficient conditions on
IFS parameters so that corresponding FIF renders a positive
curve for given positive data. The convergence of the fractal
interpolant is studied by establishing the uniform error of
rational cubic FIF to the original data defining function. Test
examples are provided to illustrate our positivity preserving
FIF scheme, tension effect and derivative analysis. It is felt
that the results are encouraging for the rational FIF treated in
this paper when compared to its classical counterpart.

2. FRACTAL INTERPOLATION FUNCTION

Let {(xi , fi ) [x1 , xn ] K, i = 1, 2, . . . , n} be a given set
of interpolation data, where K is a compact set in R, <
x1 < x2 < < xn < . Let I = [x1 , xn ], Ii = [xi , xi+1 ],
J = {1, 2, . . . , n 1} and Li : I Ii , i J be contraction
homeomorphisms such that

Construct continuous functions Fi : I K D such that

)
Fi (x1 , f1 ) = fi , Fi (xn , fn ) = fi+1 ,
|Fi (x, y) Fn (x, y )| |i ||y y |,

where x I, y, y K, and 1 < i < 1; i J. Define,

wi (x, f ) = (Li (x), Fi (x, f )) for all i J. The definition of a
FIF originates from the following proposition:
Proposition 2.1. [1] The IFS {I K; wi , i J} has a
unique fixed point G. G is the graph of a continuous function
h : I K such that h(xi ) = fi , i = 1, 2, . . . , n.
The function h in Proposition 2.1 is called a FIF corresponding to the IFS {I K; wi , i J}. Let G = {g : I
R | g is continuous, g(x1 ) = f1 and g(xn ) = fn }. Define a
metric on G by
(g1 , g2 ) = max{|g1 (x) g2 (x)| : x I} g1 , g2 G.
Then (G, ) is a complete metric space. Define the ReadBajraktarevic operator T on (G, ) as
1
T g(x) = Fi (L1
i (x), g(Li (x))), x Ii , i J .

Li (xn ) = xi+1 .

(1)

(3)

Li1 , g, Fi

Since the functions

are all continuous, obviously
T g is continuous on Ii . Also T g is continuous at each of
the points x2 , x3 , . . . , xn1 , T g(x1 ) = f1 , and T g(xn ) = fn
due to Eqns.(1)-(2). Consequently T g G. Moreover, T is a
contraction mapping on the complete metric space (G, ), i.e.,
(T f, T g) || (f, g), where || = max{|i | : i J}.
By Banach fixed point theorem, T possesses a unique fixed
point h (say) on G. This fixed point h is the FIF corresponding
to the {I K; wi , i J}. Thus it follows from Eqn. (3) that
the FIF satisfies the following functional equation:
h(Li (x)) = Fi (x, h(x)), x I, i J.

(4)

In practice, the popular FIFs are defined by the IFS:

Li (x) = ai x + bi , Fi (x, f ) = i f + si (x), i J

(5)

where i is called a vertical scaling factor of wi and =

(1 , 2 , . . . , n1 ) is the scale vector of IFS. The scale vector
gives a degree of freedom to the FIF and allows us to
modify its properties. In our construction of the rational cubic
FIF, it is assumed that si , i J are rational functions
involving four shape parameters. The following proposition
guarantee the existence of a differentiable or spline FIF. The
proof of the following proposition follows through the suitable
modification of the arguments in Barnsley et al. [2].
Proposition 2.2. Let {(xi , fi )|i = 1, 2, . . . , n} be the interpolation data with x1 < x2 < < xn . Let Li (x) = ai x + bi
satisfies Eqn. (1) and Fi (x, f ) = i f + si (x) verifies Eqn.
(x)
, pi (x), qi (x) be suitable
(2) for i J. Let si (x) = pqii(x)
chosen polynomials in x of degree M, N respectively, and
qi (x) 6= 0 x I. Suppose for r N, |i | < ari ,
i J Let Fim (x, f ) =

Li (x1 ) = xi ,

(2)

(m)

fn

(m)
sn1 (xn )
m
an1 n1

(m)

i f +si (x)
,
am
i
(m)

, where si

(m)

f1

(m)

s1 (x1 )
,
am
1 1

(x) represents the mth

(m)

derivative of si (x) with respect to x. If Fim (xn , fn ) =

(m)
m
Fi+1
(x1 , f1 ) for i = 1, 2, 3, . . . , n 2 and m = 1, 2, . . . , r,
n1
then {(Li (x), Fi (x, f ))}i=1
determines a rational FIF
r
(m)
C [x1 , xn ] and
is the rational FIF determined by
n1
{(Li (x), Fim (x, f ))}i=1
for m = 1, 2, . . . , r.

3. RATIONAL CUBIC FIF WITH FOUR FAMILIES OF SHAPE PARAMETERS

In this section we construct the rational cubic FIFs with four
families of shape parameters, which generalize a class of
classical rational cubic interpolants described by Sarfraz et
al. [11].
Theorem 3.1. Let {(xi , fi ); i = 1, 2, . . . , n} be a given data
set, where x1 < x2 < < xn and di (i = 1, 2, . . . , n)
be the derivative values at knots. Consider the IFS I
{I R; wi (x, f ) = (Li (x), Fi (x, f )), i J} where Li (x) =
ai x + bi satisfies Eqn. (1) and let Fi (x, f ) = i f + si (x),
(x)
si (x) = pqii(x)
, pi (x) are cubic polynomials and qi (x) 6= 0
are preassigned cubic polynomials with four shape
parameters
f +s (x)
and |i | < ai , i J. Let Fi1 (x, f ) = i ai i , where si (x)
represents the derivative of si (x) with respect to x. If
Fi (x1 , f1 ) = fi , Fi (xn , fn ) = fi+1 ,

(6)

Fi1 (x1 , d1 ) = di , Fi1 (xn , dn ) = di+1 ,

then the fixed point of IFS I is the graph of a C 1 -rational

cubic FIF.
Proof: Suppose = { C 1 [x1 , xn ] : (x1 ) =
f1 , (xn ) = fn , (x1 ) = d1 and (xn ) = dn }. Then (, d )
is a complete metric space, where d is the metric induced
by the C 1 -norm on C 1 [x1 , xn ]. Define the Read-Bajraktarevic
operator T on as
1
T (x) = i (L1
i (x)) + si (Li (x)), x Ii , i J.

(7)

xi+1 xi
xn x1

< 1, the condition |i | < ai < 1 gives

Since ai =
that T is a contraction operator on (, d ). From Eqn. (4)
the fixed point of of T is a fractal function that satisfies
the functional equation
(Li (x)) = i (x) + si (x), x I, i J.

(8)

The four real parameters in the rational cubic function si (x)

are evaluated using Eqn. (6) in the following way:
Substituting x = x1 and x = xn in Eqn. (8), we get,
fi = i f1 + si (x1 ), fi+1 = i fn + si (xn ), i J.

(9)

Since C 1 [x1 , xn ]. satisfies the functional equation:

(Li (x)) =

i (x) si (x)
+
, x I, i J.
ai
ai

(10)

|i |
ai

< 1, is a fractal function. Substituting x = x1

Since
and x = xn in Eqn. (10), we have two equations involving di
and di+1 respectively as
di =

s (x1 )
i dn
s (xn )
i d1
+ i
, di+1 =
+ i
, i J. (11)
ai
ai
ai
ai

The four real parameters of pi (x) are determined from Eqn.

(9) and Eqn. (11). By using similar arguments as in [1], it can
be shown that the IFS I has a unique fixed point and it is
the graph of the rational cubic FIF C 1 [x1 , xn ].

4. EVALUATION OF C1 RATIONAL CUBIC

FIF
fi
Denote i = xfi+1
, i J. For i = 1, 2, . . . , n, assume that
i+1 xi
pi (x) Pi () = Ui (1)3 +Vi (1)2 +Wi (1)2 +Zi 3 ,
qi (x) Qi () = i (1)2 +i (1)2 +i (1)2 +i 2 ,
1
, x I, i > 0, i > 0, i > 0 and i > 0;
where = xxx
n x1
i J, are shape parameters, and Ui , Vi , Wi , Zi i J are real
parameters to be evaluated. By considering Theorem 3.1 with
above pi (x) and qi (x), we consider the following functional
equation:

(Li (x)) = i (x) +

pi (x)
, x I, i J.
qi (x)

(12)

To make the rational cubic fractal function a C 1 -interpolant,

one needs to impose (xi ) = fi , (xi+1 ) = fi+1 , (xi ) =
di , (xi+1 ) = di+1 . Substituting x = x1 in Eqn. (12),
we obtain Ui = i (fi i f1 ). Similarly, x = xn gives
Zi = i (fi+1 i fn ). Now using (xi ) = di in Eqn. (12), we
get Vi = (i +i )fi +i hi di i [i (xn x1 )d1 +(i +i )f1 ].
Finally the condition (xi+1 ) = di+1 gives Wi = (i +
i )fi+1 i hi di+1 + i [i (xn x1 )dn (i + i )fn ].
Substituting the values of Ui , Vi , Wi and Zi in (13), the
rational cubic FIF is obtained as
(Li (x)) = i (x) +

pi (x)
, x I, i J,
qi (x)

(13)

where pi (x) = i (fi i f1 )(1)3 +{(i +i )fi +i hi di

i [i (xn x1 )d1 + (i + i )f1 ]}(1 )2 + {(i + i )fi+1
i hi di+1 +i [i (xn x1 )dn (i +i )fn ]}(1)2 +i (fi+1
i fn )3 , qi (x) = i (1)2 +i (1)2 +i (1)2 +i 2 .
In the absence of derivative values at the grids, we compute
them from the given data set. In this paper we use the
arithmetic mean method (see for instance [7]) for derivative
approximations for the shape preserving rational cubic FIFs:
Arithmetic mean method: The three-point difference approximation at internal grids are given as for i = 2, 3, n 1,
+hi1 i
, and at the end points of interval, the
di = hi hi1
i +hi1
2 )h1
, and
derivatives are approximated as d1 = 1 + (h1
1 +h2
(n1 n2 )hn1
dn = n1 +
.
hn1 +hn2
Example 1: We construct the rational cubic FIFs (see Figs.
1(a)-(d)) for the data {(0, 0.5), (2, 1.5), (3, 7), (9, 9), (11, 13)}
by using the following rational IFS:

{I R; wi (x, f ) = (Li (x), Fi (x, f )), i J},

(14)

Li (x) = ai x + bi , ai , bi are evaluated by using Eqn. (1), and

Pi ()
with the expression Pi () and Qi ()
Fi (x, f ) = i f + Q
i ()
as in Eqn. (13). In the construction of the rational cubic FIFs,
the choice of the scaling vectors are = (0.1, 0.1, 0.1, 0.1)
in Figs. 1(a)-(c) and = (0.001, 0.001, 0.001, 0.001) in

Fig.1(d), and i = i = 1 i = 1, 2, 3, 4 in all figures.

At the knot points, the first order derivatives are approximated
by the arithmetic mean method. By analyzing Figs. 1(a)-(c),
we conclude that when shape parameters increases the fractal
curve is tightened. In Fig. 1(d) when i is close to zero, then
the fractal curves tends to a piecewise linear interpolant for
large values of tension parameters i and i . Thus, it is verified
graphically if i 0+ and i , i simultaneously
then the rational cubic FIF modifies to the classical affine
interpolant, and hence it can be used for shape preserving
interpolants.
Remark 1: (Tension property) Eqn. (13) can be expressed
14

the C 1 -Hermite interpolant as mentioned in Remark 3. In Fig.

2(b), we take i = (0.111, 0.0901, 0.541, 0.1810) and shape
parameters as i = 100, i = 80 and i = i = 2. In Fig.
2(c), we take i = 0 and the shape parameters as in Fig.
2(b), to obtain the classical rational cubic FIF as mentioned
in Remark 2. In Fig. 2(d), we take i as in Fig. 2(b) and the
shape parameters as i = 100, i = 1 and i = i = 2 to
illustrate the effects of i . In comparison of Fig. 2(b) with
respect to Fig. 2(d), it is observed that for small values of i ,
we obtain a non-positive rational cubic FIF.
14

40

12

35

10

30

25

20

14

12

12

10

10

15

10

0
2

10

12

(a) i = i = 5 .

10

12

(b) i = i = 500 .

14

14

12

10

12

(a) i = i =
1 and i = i = 2.

10

12

(b) i = 100, i =
80 and i = i = 2.

16

80

14

70

12

60

10

12

50

8
10

10

40
6
30
4
20

2
6

10

0
4

0
0

10

(c) i = i = 5000.
Fig. 1.

12

2
4

10

12

(d) i = i = 5000.

Illustrations of shape control analysis

as (Li (x)) = i (x) + fi (1 ) + fi+1 + QiR() , where

R = i i f1 (1 )3 + ((i + i )fi + i hi di i [i (xn
x1 )d1 + (i + i )f1 ])(1 )2 + ((i + i )fi+1 i hi di+1 +
i [i (xn x1 )dn (i + i )fn ])(1 )2 i fi (1 )3
i fi+1 (1 )3 (i fi+1 + i fi )(1 )2 2 i i fn 3 . If
i 0+ and i , i simultaneously, then the rational
cubic FIF modifies to the classical affine interpolant.
Remark 2: If i = 0, i J, the rational Cubic FIF becomes
the classical rational cubic interpolation function s(x) as
3
(Li (x)) = p()
q() , where p() = i fi (1 ) + [(i + i )fi +
2
i hi di ](1 ) + [(i + i )fi+1 i hi di+1 ](1 ) 2 +
i fi+1 3 , q() = i (1)2 +i (1)2 +i (1) 2 +i 2
as described by Sarfraz and Hussain [11], (see the Fig. 2(c)).
Remark 3: It is interesting to note that when i = i = 1, i =
i = 2 and i = 0 for i J in Eqn. (12), we obtain
(x) = (2 3 3 2 + 1)fi + ( 3 2 2 + )di hi + (2 3 +
xxi
3 2 )fi+1 + ( 3 2 )di+1 hi , = xi+1
xi , x [xi , xi+1 ].
Hence is the C 1 -Hermite interpolant over I.
Example 2: By taking the same data as in Example 1, we want
to show the effect of the scaling factor i and shape parameters
in the rational cubic FIF. In Fig. 2(a), we take i = 0 and
shape parameters as i = i = 1 and i = i = 2. we obtain

10

(c) i = 100, i =
80 and i = i = 2 .
Fig. 2.

12

10

10

12

(d) i = 100, i =
1 and i = i = 2.

The rational cubic FIF with arbitrary parameter

5. CONVERGENCE PROPERTIES OF RATIONAL CUBIC FIF

In the following theorem, the uniform error bound for a
rational cubic FIF (x) is obtained. It can be proved using
the definition of the Read-Bajraktarevic operators for which
is a fixed point and by applying the Mean Value Theorem.
We provide a sketch of the proof.
Theorem 5.1. Let and s be the rational cubic FIF and
classical rational cubic interpolant with respect to the data
{(xi , fi ), i =1,2,...,n} generated from an original function f
C 3 [x1 , xn ]. Let di , i = 1, 2, ..., n be the bounded derivatives
at the knots. Suppose |i | < ai , || = max{|i | : i J} ,
h = max{|hi | : i J}, c = max{|ci | : i J}, and the shape
parameters i , i , i and i , i J, for
some 0 < < 1. Then
||
(ksk + K) + 21 kf (3) k h3 c
k f k 1||

Proof: Suppose that the rational cubic polynomial

ri (i , x) involved in the IFS that generate rational cubic FIF
(i ,x)
satisfies | ri
| K for |i | (0, kai ) ,|i | < k < 1,
i

x Ii = [xi , xi+1 ], i J. Then, from Lemma 4.1 of [3]

|i |
(ksk + K).
k sk 1|
i |
Also it is known from Theorem 5 of [11], the uniform error
bound of the rational cubic function for f (x) C 3 [x1 , xn ],
kf sk 12 kf (3) k max {h3i ci } for some suitable
1in1

constant ci independent of hi .
Using both above results in kf k ksk +ksf k,
|i |
k f k 1|
(ksk + K) + 12 kf (3) k h3 c,
i |
we have obtained the desired bound for k f k . Since
h
|| xn x
, the above theorem establishes that k
1
f k = O(h). That is, converges to f uniformly as
mesh norm approaches zero. Further if we elect i such that
h3i
|i | < a3i = (xn x
i J, then converges to f
1)
with O(h3 ) which is similar to the convergence result of its
classical nonrecursive counterpart s.

6. POSITIVITY PRESERVING FRACTAL INTERPOLANT

For an arbitrary selection of the scaling factors and shape
parameters, the rational cubic FIF described above may not
be positive, if the data set is positive. This is very similar to
the ordinary spline schemes that do not provide the desired
shape features of data. We have to restrict the scaling factor
and shape parameters for a positive preserving rational cubic
FIF as described in the following theorem:
Theorem 6.1. Let {(xi , fi , di ), i = 1, 2, . . . , n} be a given
positive data. If (i) the scaling factor i , is selected as
fi fi+1
,
}), i J,
(15)
f1 fn
(ii) the shape parameters i , i , i and i , are restricted as
hi d
i > 0, i > 0, i max{0, i [1 + i ]},
mi
(16)
hi di+1
]},
i

J,
i max{0, i [1
ni
i [0, min{ai ,

where mi = fi i f1 , ni = fi+1 i fn , di = di ai di 1 ,
di+1 = di+1 iadi n , then we obtain a positivity preserving
C 1 -rational cubic FIF .
Pi ()
1
, = xxx
,
Proof: We have (Li (x)) = i (x) + Q
i ()
n x1
x I. If (x) 0 it is easy to verify that for i J, when
i 0, i J, the sufficient condition for (Li (x)) > 0 for
Pi ()
> 0 [0, 1].
all x I is Q
i ()
So the initial condition on the scaling factors are i 0, i J.
Since Qi () > 0 [0, 1], i > 0, i > 0, i > 0
Pi ()
and i > 0; i J, the positivity of rational function Q
i ()
depends on the positivity of Pi (). For our convenience, we
write Pi () as
Pi () = Ui (1 )3 + Vi (1 )2 + Wi (1 )2 + Zi 3 . Now
Pi () > 0 if Ui > 0, Vi > 0, Wi > 0, Zi > 0 (cf. Section 4).
Now Ui > 0, Zi > 0 if Eqn. (15) is true. Also Vi > 0, Wi > 0
when Eqn. (16) is valid. Since (x) is constructed iteratively
(Li (.)) 0 (.) 0 implies (x) 0; x I.

Consequence. Setting i = 0 in Eqn. (17), we obtain

i > 0, i > 0, i max{0, i [1 + hfi di i ]}, i
i+1
max{0, i [1 hfi di+1
]}, that provide a set of sufficient
conditions for the positivity of rational cubic spline s. It
is worthwhile to mention that these conditions on shape
parameters are weaker than those obtained by Sarfraz and
Hussain [11].

7. ILLUSTRATION OF POSITIVE PRESERVING RESULTS

Consider the positive data set {(0, 0.5), (2, 1.5), (3, 7),
(9, 9), (11, 13)}. If the selection of i is not according to
Eqn. (15), then we may obtain a non-positive rational cubic
FIF. For instance the selection of 1 is not according to Eqn.
(15), and hence a part of the rational cubic FIF in Fig. 3(a)
related to the first sub-interval is lying in the fourth quadrant.
Using the rational IFS Eqn. (14), we have constructed all our
positivity preserving rational cubic FIFs, where the derivative
values are approximated by the arithmetic mean method as
d1 = 2.8333, d2 = 3.8333, d3 = 4.7619, d4 = 0, and d5 =
2.4167. In order to carry the positive nature of above positive
data by our rational cubic FIF, the IFS parameters are chosen
with respect to Eqns. (15)-(16), respectively. The restriction of
the scaling factors prescribed by Eqn. (15) are: 1 [0, 0.115),
2 [0, 0.0909), 3 [0, 0.545), 4 [0, 0.181). A standard
rational cubic positive FIF in Fig. 3(b) is constructed with
scaling vector as = (0.111, 0.0901, 0.541, 0.1810) and shape
parameters as = (0.5, 0.6, 0.7, 0.8), = (0.2, 0.3, 0.4, 0.5),
= (6, 0.2, 0.3, 0.4) and = (17, 0.1, 0.2, 0.3). The scaling
factors are the key parameters of our rational cubic FIF.
Theoretically due to the implicit and recursive nature of the
fractal interpolant, a change in a particular i value may
be rippled through the entire configuration. To study this in
practice, we change a particular value of i and observe
its effect on the shape of Figs. 3(c)-(f) are constructed by
changing the scaling factor 1 , 2 , 3 and 4 as 0.001 with
respect to other IFS parameter of Fig. 3(b), respectively. By
analyzing Fig. 3(c) with respect to Fig. 3(b), we observe major
changes in [0, 2] and minute changes in other intervals. The
effects of 2 and 4 (see Figs. 3(d),3(f)) are similar to 1 , and
all these parameter can be taken as moderately local in nature.
The effect of 3 is global in nature (see Fig. 3(e)). Thus the
effect of perturbation of i may be global or local in nature.
he user can get the desired positive rational cubic FIF by
playing with the scaling factors (within the restricted interval)
and shape parameters through some optimization technique.
In Fig. 4, we have showed graphically that the derivative of
a rational cubic FIF are irregular but in the classical case, it
is smooth except at knot points. The derivatives of rational
cubic FIFs as described in Figs. 3(b)-(e) are plotted in Figs.
4(a)-(d). The fractality in the derivative of the rational cubic
FIF can be controlled over any subinterval with proper choice
of the scaling factor.

40

40

35

35

30

30

40

40

30

30
25

25

20

20
20
20

10

10

15
15
10

10

10

20

20

10
5
5

0
5

0
0

10

12

10

12

40

(b) Standard rational

cubic positive FIF .

(a) Non-positive rational

cubic FIF.
40

40

35

35

30

30

25

25

30

30

40

10

12

(a) Standard Fractal

derivative.

10

12

(b) Effects of 1 in Fig.

4(a).

20

40
35

20

20

15

15

10

10

10
30

25
20

10
5
0

15

10

12

10

20
0

10

12
5

30

(c) Effects of 1 in Fig.

3(b).
16

40

14

35

12

30

10

25

20

15

10

10

(e) Effects of 3 in Fig.

3(b).
Fig. 3.

(d) Effects of 2 in Fig.

3(b).

12

40

10

12

(e) Effects of 2 in Fig.

4(a).
Fig. 4.

10

12

(f) Effects of 3 in Fig.

4(a).

Derivative of rational cubic FIF.

9. ACKNOWLEDGEMENT
0

10

12

(f) Effects of 4 in Fig.

3(b).

Rational cubic FIF and its positivity aspects

8. CONCLUSION
We have developed a C 1 -rational cubic FIFs that contain four
families of shape parameters. With a zero scaling vector,
the developed rational cubic FIF reduces to the existing
classical rational cubic interpolant with four families of
shape parameters. A uniform error bound has been developed
between the rational cubic FIF and an original function, from
which it is deduced that the rational cubic FIF converges
uniformly to the original function. Data-dependent sufficient
constraints have been developed on the scaling factors. Two
of the free parameters are used to preserve the positivity of
the data, while the other two free parameters are used to refine
the shape of curve. The power of the proposed rational cubic
FIF have been demonstrated through suitable examples. The
present interpolation scheme may find applications in areas
such as tomography, computer graphics, CAGD, animation,
visual space simulation, VLSI and image processing.

The partial support of the Department of Science and Technology of Govt. of India (SERC DST Project No. SR/S4/MS:
694/10) is gratefully acknowledged.

References
[1] Barnsley, M. F.: Fractal functions and interpolations. Constr. Approx.,
303-329, 1986.
[2] Barnsley, M. F., Harrington, A. N.: The calculus of fractal interpolation
functions. J. Approx. Theory., 14-34, 1989.
[3] Chand, A. K. B., Kapoor, G. P.: Generalized cubic spline fractal
interpolation functions. SIAM J. Numer. Anal., 655-676, 2006.
[4] Chand, A. K. B., Viswanathan, P.: Cubic Hermite and spline fractal
interpolation functions. AIP conference Proceedings. 1479, 1467-1470,
2012.
[5] Chand, A. K. B. Vijender, N., Navascues, M. A.: Shape preservation of scientific data through rational fractal splines. Calcolo DOI10
1007/s10092-013-0088-2.
[6] Delbourgo, R., Gregory, J. A.: Shape preserving piecewise rational
interpolation. SIAM J. Sci. Stat. Comp. 6(1), 967-976, 1985.
[7] Delbourgo, R., Gregory, J. A.: C 2 rational quadratic spline interpolation
to monotonic data. IMA J. Numer. Anal. 3, 141-152, 1983.
[8] Gregory, J. A., Sarfraz, M.: A rational spline with tension. CAGD 7,
1-13, 1990.
[9] Kumar, V., Gavrilova, L., Tan, C. J. K., Ecuyer, P.L.: Computational
science and its applications. In: ICCSA. Springer, Berlin, 2003.
[10] Sarfraz, M., Al-Mulhem, M., Ashraf, F.: Preserving monotonic shape of
the data using piecewise rational cubic functions. Comput. Graph. 21(1),
5-14, 1997.
[11] Sarfraz, M., Hussain, M. Z., Hussain, M.:Shape-preserving curve interpolation. J. Comp.Math., 35-53, 2012.
[12] Shrivastava, M., Joseph, J.: C 2 -rational cubic spline involving tension
parameters. Proc. Indian Acad. Sci. Math. Sci. 110(3), 305-314, 2000.
[13] Tian, M.: Monotonicity preserving piecewise rational cubic interpolation. Int. J. Math. Anal. 5, 99-104, 2011.