Applied Mathematics and Computation 172 (2006) 91–100

www.elsevier.com/locate/amc

The calculation of integrals involving

B-splines by means of recursion relations
M. Bhatti a, P. Bracken b,*

a
Department of Physics and Geology, University of Texas Pan American,
Edinburg, TX 78541-2999, USA
b
Department of Mathematics, Univertsity of Texas Pan American, Edinburg, TX 78541-2999, USA

Abstract

A procedure is given for constructing the exact integrals involving B-splines using
recursion relations. The recursive integrals form the basic constituents for the exact eval-
uation of integrals that appear in the calculation of atomic and molecular properties.
The method can be applied to solve differential equations as well as to produce a com-
plete set of basis functions that may approximate a function arbitrarily well depending
on the degree k and the number of B-splines that are employed in the approximation.
The advantage of this method is that the recursions developed over a fixed interval rep-
resent exact results for the particular integral involved. Several examples are also pro-
vided to show how these recursion relations can be applied to evaluate integrals
involving multiple B-splines of same or different degree. Closed forms of these integrals
involve complicated recursion relations.
Ó 2005 Elsevier Inc. All rights reserved.

Keywords: B-spline; Atomic and molecular integrals; Recursion relations; Many body theory

*
Corresponding author.
E-mail address: bracken@panam.edu (P. Bracken).

0096-3003/$ - see front matter Ó 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2005.01.119
92 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100

1. Introduction

The precision calculation of molecular integrals is extremely important for

applications to such diverse areas as statistical mechanics and quantum chem-
istry, and constitutes a fundamental and important part of the theoretical
studies of atomic and molecular systems. Improvement of the current compu-
tational schemes and understanding of these methods in better ways is impor-
tant to the further development of the computational studies of molecular
systems. Molecular integrals appear in applications such as many body pertur-
bation theory applied to molecular systems, as well as the calculation of differ-
ent atomic and molecular properties. Molecular integrals typically form the
basic constituents for the exact calculation of these properties. Integrals fre-
quently appear in calculations of interaction energies between atoms in mole-
cules and between molecules themselves.
One property of physical interest to which the results presented here could
be applied is to the calculation of the polarizability and electronic Coulomb
integrals (RK-integrals) [1–3]. Polarizability is defined as the response of a sys-
tem to an external electric field. The study of polarizability of the excited states
of the hydrogen atom is of particular interest in the field of atomic physics to
check the accuracy of different computational techniques, for example, the cou-
pled Hartree–Fock method or the finite perturbation method [4,5]. The solu-
tion of the differential Schrödinger equation relating to the polarizability of
excited states become unsolvable analytically for complex atomic systems,
and in this instance, approximate methods are essential and involve integrals
to calculate matrix elements [6]. The procedure that will be discussed here
makes use of the B-spline method. This method creates a complete set of basis
functions that will approximate a function arbitrarily well depending on the
degree k and number N of B-splines that are employed in the approximation.
A function
PN f(x) can be expressed in terms of an expansion of the form
f ðxÞ ¼ i¼1 C i Bki ðxÞ, where k represents the degree of the B-splines, and Ci is
the ith coefficient of the series. The purpose of this paper then is to develop sev-
eral recursion relations for various types of integrals which involve B-splines
and appear in these approximate, yet high precision, type of solutions. The
advantage to this method is that the recursions developed over a fixed interval
represent exact results for the particular integral involved. A number of exam-
ples are also included to show how they can be applied.

2. B-splines and integrals

Let us first briefly review the idea of B-splines and a working definition of
them [7,8]. The principal use of the B-splines Bki (i = 0, ±1, ±2, . . .) is as a basis
for the set of all kth degree splines that have the same knot sequence. Suppose
 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100 93

that an infinite set of knots {ti}, or if the interval is finite, a finite set of knots,
some of which may be identical at the end points, has been prescribed in such a
way that tj < tj+1. The B-splines are a function of these knots and are defined
recursively as follows. The B-splines of degree zero are defined by B0i ðxÞ ¼ 1
when ti 6 x 6 ti+1 and B0i ðxÞ ¼ 0 otherwise. For fixed x, there is a unique inte-
ger m such that tm 6 x < tm+1 and so
X1
B0i ðxÞ ¼ B0m ðxÞ ¼ 1:
i¼
1

Initializing the sequence of B-splines with the functions B0i ðxÞ, all the higher
degree B-splines are given by a simple recursive definition, namely,



k x
ti k
1 tiþkþ1
x
Bi ðxÞ ¼ B ðxÞ þ Bk
1 ðxÞ; ð1Þ
tiþk
ti i tiþkþ1
tiþ1 iþ1

for k P 1 where k = 1, 2, . . . and i = 0, ±1, ±2, . . .. The ti are a fixed set of

numbers on or in a subinterval of the real line. For example, we have that
for k = 1



1 x
ti 0 tiþ2
x
Bi ðxÞ ¼ B ðxÞ þ B0 ðxÞ: ð2Þ
tiþ1
ti i tiþ2
tiþ1 iþ1
Obviously, B-splines can be differentiated and integrated, and since we will
make use of some of the basic results of these operations in what follows, we
will mention the two most important results here. The derivative of a B-spline
is given by



d k k k
Bi ðxÞ ¼ Bk
1 ðxÞ
Bk
1 ðxÞ: ð3Þ
dx tiþk
ti i tiþkþ1
tiþ1 iþ1
This can be established by induction using (1). From the following differen-
tiation formula


d X 1
kþ1
X1
ci
ci
1
ci B ðxÞ ¼ ðk þ 1Þ Bk ðxÞ;
dx i¼
1 i i¼
1
tiþkþ1
ti i
then by selecting cj = 0 for j = i
1, i
2, . . . and ci = ci+1 = ci+2 =    = 1, we
find that
d X1
kþ1
Bkþ1 ðxÞ ¼ Bk ðxÞ: ð4Þ
dx j¼i j tiþkþ1
ti i

Integrating both sides of (4), we have the basic result for the integration of a
B-spline
Z x
tiþkþ1
ti X
1
Bki ðsÞ ds ¼ Bkþ1 ðxÞ: ð5Þ

1 k þ 1 j¼i j
94 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100

At this point, several recursion relations which involve integrals of B-splines

will be formulated. We work on a finite interval from zero to x, taking all the
knots to the left of the origin to be zero. The last distinct knot on the right will
be the endpoint of the interval and all further knots to the right will equal that
one. In other words, multiplicity of the knots will be k+1 at the extreme points
in the sequence. The lower limit on the integral will be zero, and let us intro-
duce a notation for the integral of Bki ðsÞ as follows:
Z x
tiþkþ1
ti X
1
I 0i ðk; xÞ ¼ Bki ðsÞ ds ¼ Bkþ1 ðxÞ: ð6Þ
0 k þ 1 j¼i j

This will be used to initialize a recursion relation for the set of integrals de-
fined by
Z x
I pi ðk; xÞ ¼ sp Bki ðsÞ ds: ð7Þ
0

All of the integrals I pi ðk; xÞ depend on the upper limit of integration x. Inte-
grating by parts once for the case in which p = 1 leads to the expression
Z x
tiþkþ1
ti X
1
tiþkþ1
ti X
1
tjþkþ2
tj
sBki ðsÞ ds ¼ x Bkþ1
j ðxÞ

0 k þ 1 j¼i k þ 1 j¼i kþ2
X
1
 Bkþ2
l ðxÞ
l¼j
" #
tiþkþ1
ti X 1 X1
¼ x Bkþ1
j ðxÞ
I 0j ðk þ 1; xÞ :
kþ1 j¼i j¼i

One of the advantages of this approach is that it is possible to obtain exact

results for the integrals as a function of x, however to do so, it is necessary to
be able to specify the interval (ti, ti+1) in which the value of x exists, since the B-
splines are defined piecewise over each (ti, ti + 1). Applying integration by parts
recursively we obtain that,
Z x
I pi ðk; xÞ ¼ sp Bki ðsÞ ds
0
1 Z x
tiþkþ1
ti X
1
tiþkþ1
ti X
¼ xp Bkþ1 ðxÞ
p sp
1 Bkþ1
j ðsÞ ds:
k þ 1 j¼i j k þ 1 j¼i 0

Using the notation above, this can be written in the form of a recursion rela-
tion, which we summarize in the following statement.
 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100 95

Proposition 1. The integrals I pi ðk; xÞ satisfy the recursion relation

tiþkþ1
ti X
1
I pi ðk; xÞ ¼ xp I 0i ðk; xÞ
p I p
1 ðk þ 1; xÞ: ð8Þ
k þ 1 j¼i j
Another class of integral that is useful since it appears in frequent applica-
tions of the type of calculation discussed in the introduction and is amenable to
being expressed in the form of a recursion relation in the same sort of way as
(8) is the following integral (Table 1)
Z x
J ðk 1 ; k 2 ; i; j; xÞ ¼ Bki 1 ðsÞBkj 2 ðsÞ ds: ð9Þ
0

These integrals J clearly have the symmetry property

J ðk 1 ; k 2 ; i; j; xÞ ¼ J ðk 2 ; k 1 ; j; i; xÞ ð10Þ
Applying integration by parts and using the differentiation and integration
formulas (3) and (5), apply integration by parts once, we obtain
Z x
tjþk2 þ1
tj k1 X 1
tjþk2 þ1
tj k1
Bki 1 ðsÞBkj 2 ðsÞ ds ¼ Bi ðxÞ Bkq2 þ1 ðxÞ

0 k 2 þ 1 q¼j
k 2 þ 1 t iþk 1
t i

X 1 Z x
tjþk2 þ1
tj
 Bqk2 þ1 ðsÞBki 1
1 ðsÞ ds þ
q¼j 0 k2 þ 1
X1 Z x
k1 k 1
1
 Bk2 þ1 ðsÞBiþ1 ðsÞ ds:
tiþk1 þ1
tiþ1 q¼j 0 q

Introducing the notation defined in Eqs. (6) and (10) into the expression
above, we obtain the following proposition.

Proposition 2. The integrals defined in Eq. (9) satisfy the following recursion
relation
tjþk2 þ1
tj k1
J ðk 1 ; k 2 ; i; j; xÞ ¼ Bki 1 ðxÞI 0j ðk 2 ; xÞ

k 2 þ 1 tiþk1
ti
X 1
tjþk2 þ1
tj k1
 J ðk 1
1; k 2 þ 1; i; q; xÞ þ
q¼j
k 2 þ 1 t iþk 1 þ1
tiþ1
X 1
 J ðk 1
1; k 2 þ 1; i þ 1; q; xÞ: ð11Þ
q¼j

Finally, a mixed recursion relation can be written down for the complicated
integral defined by
Z x
J ðpÞ ðk 1 ; k 2 ; i; j; xÞ ¼ sp Bki 1 ðsÞBkj 2 ðsÞ ds: ð12Þ
0
96 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100

Table 1
Integrals calculated using (7) and (8)
I 0i ðk; xÞ, i = 1, 2, . . ., k + 1 and k = 1, 2, 3, 4, 5
x2 7x4 3x5
I 0
1 ð1; xÞ x
I 0
3 ð4; xÞ 2x2
3x3 þ

2 4 8
0 x2 11x4 17x5
I 0 ð1; xÞ I 0
2 ð4; xÞ 3
x
þ
2 12 72
0 x3 x 4
13x 5
I
2 ð2; xÞ x
x2 þ I 0
1 ð4; xÞ

3 6 180
3 5
x x
I 0
1 ð2; xÞ x2
I 00 ð4; xÞ
2 90
x3 5x2 10x3 5x4 x6
0
I 0 ð2; xÞ I 0
5 ð5; xÞ x
þ
þ x5

6 2 3 2 6
3x2 x4 5x2 35x4 15x5 31x6
0
I
3 ð3; xÞ x
þ x3
I 0
4 ð5; xÞ 3

5x þ
þ
2 4 2 8 8 96
2 3 4 3 4 5
3x 3x 7x 5x 55x 85x 575x6
I 0
2 ð3; xÞ
þ I 0
3 ð5; xÞ
þ

2 2 16 3 24 72 2592
0 x3 11x4 5x4 13x5 37x6
I
1 ð3; xÞ
I 0
2 ð5; xÞ
þ
2 48 12 36 432
4 5 6
0 x x 5x
I 0 ð3; xÞ I 0
1 ð5; xÞ

24 18 216
x5 x 6
I 0
4 ð4; xÞ x
2x2 þ 2x3
x4 þ I 00 ð5; xÞ
5 324
I 1i ðk; xÞ i = 1, 2, . . ., k + 1 and k = 1, 2, 3, 4
x2 x3 3x4 11x5
I 1
1 ð1; xÞ
I 13 ð3; xÞ

2
1 8 60
3 5
x x
I 10 ð1; xÞ I 10 ð3; xÞ
3 30
x2 2x3 x4 x2 4x3 3x4 4x5 x6
I 1
2 ð2; xÞ
þ I 1
4 ð4; xÞ
þ
þ
2 3 4 2 3 2 5 6
2x3 3x4 4x3 9x4 7x5 5x6
I 1
1 ð2; xÞ
I 1
3 ð4; xÞ
þ þ
3 8 3 4 5 16
4 4 5
x 3x 11x 85x6
I 10 ð2; xÞ I 1
2 ð4; xÞ
þ
8 4 15 432
x2 3x4 x5 2x 5
13x 6
I 1
3 ð3; xÞ
x3 þ
I 1
1 ð4; xÞ

2 4 5 15 216
9x4 7x5 x6
I 1
2 ð3; xÞ 3
x
þ I 10 ð4; xÞ
8 20 108
I 2i ðk; xÞ p = 2, i = 1, 2, . . ., k + 1 and k = 1, 2, 3
x3 x4
I 2
1 ð1; xÞ

3 4
x4
I 20 ð1; xÞ
4
x3 x4 x5
I 2
2 ð2; xÞ
þ
3 2 5
4
x 3x5
I 2
1 ð2; xÞ

2 10
 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100 97

Table 1 (continued)
I 20 ð2; xÞ x5
10
x3 3x4 3x5 x6
I 2
3 ð3; xÞ
þ

3 4 5 6
3x4 9x5 7x6
I 2
2 ð3; xÞ
þ
4 10 24
5 6
3x 11x
I 2
1 ð3; xÞ

10 72
6
x
I 20 ð3; xÞ
36
I 3i ðk; xÞ i = 1, 2, . . ., k + 1 and k = 1, 2
x4 x5
I 3
1 ð1; xÞ

4 5
x5
I 30 ð1; xÞ
5
x4 2x5 x6
I 3
2 ð2; xÞ
þ
4 5 6
5 6
2x x
I 3
1 ð2; xÞ

5 4
x6
I 30 ð2; xÞ
12
I 4i ðk; xÞ, i = 1, 2, . . ., k + 1 and k = 1
x5 x6
I 4
1 ð1; xÞ

5 6
6
x
I 40 ð1; xÞ
6
The integrals I pi ðk; xÞ are evaluated for variety of i and k values assuming x in the interval (0, 1).

Thus, when p = 0, we have J(0)  J, and after one application of integration

by parts, we have
J ðpÞ ðk 1 ; k 2 ; i; j; xÞ
tk þiþ1
ti p X 1
tk þiþ1
ti k2
¼ 1 x Bqk1 þ1 ðxÞBkj 2 ðxÞ
1 xp
k1 þ 1 q¼i
k 1 þ 1 t jþk 2

t j
X 1 Z x
tkþiþ1
ti k2
 Bq ðrÞBkj 2
1 ðrÞ dr þ
k 1 þ1
xp
q¼i 0 k 1 þ 1 t jþk 2 þ1
t jþ1
"

X 1 Z x Z x
t k 1 þiþ1
t i
 Bqk1 þ1 ðrÞBkjþ1 2
ðrÞ dr
p sp
1
q¼i 0 0 k1 þ 1
X1 X 1 Z s
tk1 þiþ1
ti k2
 B
q¼i q
k 1 þ1
ðsÞB k2
j ðsÞ
Bk1 þ1 ðrÞBkj 2
1 ðrÞ dr
k 1 þ 1 tjþk2
tj q¼i 0 q
#
X 1 Z s
tk1 þiþ1
ti k2 k 1 þ1 k2
þ B ðrÞBjþ1 ðrÞ dr ds:
k 1 þ 1 tjþk2 þ1
tjþ1 q¼i 0 q
98 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100

Let us now introduce the notation defined in Eqs. (6), (9) and (12) into this
expression to obtain the following claim.

Proposition 3. The integral defined by (12) has the recursive representation given
by
tk þiþ1
ti k2
J ðpÞ ðk 1 ; k 2 ; i; j; xÞ ¼ xp I 0i ðk 1 ; xÞBkj 2 ðxÞ
1 xp
k 1 þ 1 tjþk2
tj
X 1
tk þiþ1
ti k2
 J ðk 1 þ 1; k 2
1; q; j; xÞ þ 1 xp
q¼i
k 1 þ 1 t jþk 2 þ1
t jþ1
X 1
tk þiþ1
ti
 J ðk 1 þ 1; k 2
1; q; j þ 1; xÞ
p 1
q¼i
k1 þ 1
X 1
tk þiþ1
ti k2
 J ðp
1Þ ðk 1 þ 1; k 2 ; q; j; xÞ þ p 1
q¼i
k 1 þ 1 t jþk 2
t j
X 1 Z x
tk þiþ1
ti
 sp
1 J ðk 1 þ 1; k 2
1; q; j; sÞ ds
p 1
q¼i 0 k1 þ 1
X 1 Z x
k2
 sp
1 J ðk 1 þ 1; k 2
1; q; j þ 1; sÞ ds;
tjþk2 þ1
tjþ1 q¼i 0

where the integrals J are defined in ð9Þ.

Finally, a horizontal recursion relation can be developed based on the result
in Proposition 3 in the following way. Integrating by parts once, we can write
Z x Z
1 p x k1
p
1
s J ðk 1 ; k 2
1; q; r; sÞ ds ¼ x Bq ðrÞBkr 2
1 ðrÞ dr
0 p 0
Z
1 x p k1

s Bq ðsÞBrk2
1 ðsÞ ds:
p 0
Using this to replace the last two integrals in (13), several cancelations take
place and we are left with the following result.

Proposition 4. The integral defined by (12) gives rise to a horizontal recursion

relation of the form
tk þiþ1
ti
J ðpÞ ðk 1 ; k 2 ; i; j; xÞ ¼ xp I 0i ðk 1 ; xÞBkj 2 ðxÞ
p 1
k1 þ 1
X 1
tk þiþ1
ti k2
 J ðp
1Þ ðk 1 þ 1; k 2 ; q; j; xÞ
1
q¼i
k 1 þ 1 tk2 þj
tj
X
1
tk1 þiþ1
ti
 J ðpÞ ðk 1 þ 1; k 2
1; q; j; xÞ þ
q¼i
k1 þ 1
k2 X1
 J ðpÞ ðk 1 þ 1; k 2
1; q; j þ 1; xÞ:
tk2 þjþ1
tjþ1 q¼i
 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100 99

3. Some simple examples

As an example which makes use of (11), consider the case in which the dis-
tinct knots are taken at the points t0 = 0, t1 = 1, t2 = 2 and t3 = 3, so x is in the
interval [0, 3]. Additional knots are taken equal to the values of t0 and t3 as the
degree of the polynomial increases. Thus t
1 = 0 and t4 = 3 are used to deter-
mine the B2j ðxÞ. If we apply (2), we obtain

B1
1 ðxÞ ¼ ð1
xÞB00 ðxÞ; B10 ðxÞ ¼ xB00 ðxÞ þ ð2
xÞB01 ðxÞ;
B11 ðxÞ ¼ ðx
1ÞB01 ðxÞ þ ð3
xÞB02 ðxÞ; B12 ðxÞ ¼ ðx
2ÞB02 ðxÞ:

Higher order B-splines can be calculated recursively from these by

applying (1). The integral J(k1,k2;i,j;x) will be determined using (11) when
0 < x < 1 for several different values of the arguments. When k1 = 1, and
k2 = 0, we have
Z X
x 1
tjþ1
tj X1
B1i ðsÞB0j ðsÞ ds ¼ ðtjþ1
tj ÞB1i ðxÞ B1q ðxÞ
J ð0; 1; i; q; xÞ
0 q¼j
tiþ1
ti q¼j
tjþ1
tj X 1
þ J ð1; 1; q; i þ 1; xÞ:
tiþ2
tiþ1 q¼j

When j = 0 and i = 0, the only integral which survives on the right is

J(1, 0; 0, 0; x) = x2/2, and when i = 1, j = 1, all of the integrals J on the right van-
ish when 0 < x < 1. Therefore, using the values for the knots ti,
Z x Z x
x2 x2
B10 ðsÞB00 ðsÞ ds ¼ x2
¼ ; B11 ðsÞB01 ðsÞ ds ¼ 0:
0 2 2 0

It can be verified by direct calculation in these cases that the left hand side
matches the right side here. Similarly, when k1 = 1 and k2 = 1
Z x
tjþ1
tj 1 X 1
tjþ2
tj X 1
B1i ðsÞB1j ðsÞ ds ¼ Bi ðxÞ B2q ðxÞ
J ð2; 0; q; i; xÞ
0 2 q¼j
2ðtiþ1
ti Þ q¼j
tjþ2
tj X 1
þ J ð2; 0; q; i þ 1; xÞ:
2ðtiþ2
tiþ1 Þ q¼j

When i = 0, j = 0 only J(2, 0; 0, 0; x) = x3/6 contributes on the right when

0 < x < 1 and for k1 = 1, k2 = 2, i = j = 0 only J(3, 0; 0, 0; x) = x4/24 and we
obtain
Z x Z x
x3 x4
B10 ðsÞB10 ðsÞ ds ¼ ; B10 ðsÞB20 ðsÞ ds ¼ :
0 3 0 8
100 M. Bhatti, P. Bracken / Appl. Math. Comput. 172 (2006) 91–100

References

[1] W.R. Johnson, S.A. Blundell, J. Sapirstein, Phys. Rev. A37 (1988) 307.
[2] W.R. Johnson, M. Idrees, J. Sapirstein, Phys. Rev. A37 (1987) 378.
[3] D.M. Bishop, J. Pipen, Chem. Phys. Lett. 236 (1995) 15.
[4] T.M. Miller, B. Bederson, Phys. Rev. A14 (1976) 1572.
[5] T.M. Miller, B. Bederson, Adv. Atom. Mol. Phys. 13 (1977) 1.
[6] M. Bhatti, K. Coleman, W. Perger, Phys. Rev. A68 (2003) 044503-1.
[7] W. Cheney, D. Kincaid, Numerical Mathematics and Computing, fourth ed., Brooks/Cole,
1999.
[8] C. De Boor, A Practical Guide to Splines, second ed., Springer-Verlag, New York, 2001.