Professional Documents
Culture Documents
and Computational
Applications
Article
Dual Taylor Series, Spline Based Function and
Integral Approximation and Applications
Roy M. Howard
School of Electrical Engineering, Computing and Mathematical Sciences, Curtin University, GPO Box U1987,
Perth 6845, Australia; r.howard@curtin.edu.au
Received: 2 March 2019; Accepted: 25 March 2019; Published: 1 April 2019
Abstract: In this paper, function approximation is utilized to establish functional series approximations
to integrals. The starting point is the definition of a dual Taylor series, which is a natural extension
of a Taylor series, and spline based series approximation. It is shown that a spline based series
approximation to an integral yields, in general, a higher accuracy for a set order of approximation
than a dual Taylor series, a Taylor series and an antiderivative series. A spline based series for an
integral has many applications and indicative examples are detailed. These include a series for the
exponential function, which coincides with a Padé series, new series for the logarithm function as
well as new series for integral defined functions such as the Fresnel Sine integral function. It is shown
that these series are more accurate and have larger regions of convergence than corresponding Taylor
series. The spline based series for an integral can be used to define algorithms for highly accurate
approximations for the logarithm function, the exponential function, rational numbers to a fractional
power and the inverse sine, inverse cosine and inverse tangent functions. These algorithms are used
to establish highly accurate approximations for π and Catalan’s constant. The use of sub-intervals
allows the region of convergence for an integral approximation to be extended.
Keywords: integral approximation; function approximation; Taylor series; dual Taylor series; spline
approximation; antiderivative series; exponential function; logarithm function; Fresnel sine integral;
Padé series; Catalan’s constant
1. Introduction
The mathematics underpinning dynamic behaviour is of fundamental importance to modern
science and technology and integration theory is foundational. The history of integration dates from
early recorded history, with area calculations being prominent, e.g., [1,2]. The modern approach to
integration commences with Newton and Leibniz and Thomson [3,4] provides a lucid, and up to
date, perspective on the approaches of Newton, Riemann and Lebesgue and the more recent work of
Henstock and Kurzweil.
A useful starting point for integration theory is the second part of the Fundamental Theorem of
Calculus which states
Zβ
f (t)dt = F(β) − F(α), (1)
α
where f = F(1) on [α, β] for some antiderivative function F, assuming f is integrable, e.g., [4–6].
However, within all frameworks of integration, a practical problem is to determine antiderivative
functions for, or suitable analytical approximations to, specified integrals. Despite the impressive
collection of results that can be found in tables books such as Gradsteyn and Ryzhik [7], the problem of
Zt
I (t) = f (λ)dλ, f ∈ =, t ∈ [α, β]. (2)
α
Approaches include use of integration by parts, e.g., [8], use of Taylor series, asymptotic expansions,
e.g., [9,10], etc.
A useful approximation approach is to use uniform convergence, bounded convergence, monotone
convergence or dominated convergence of a sequence of functions and a representative statement
arising from dominated convergence, e.g., [11], is: If fi is a sequence of Lebesgue integrable functions
on [α, β], lim fi (t) = f (t) pointwise almost everywhere on [α, β], and there exists a Lebesgue integrable
i→∞
function g on [α, β] with the property fi (t) < g(t), t ∈ [α, β] almost everywhere and for all i, then
Zt Zt
lim fi (λ)dλ = f (λ)dλ = lim Fi (t), t ∈ [α, β], (3)
i→∞ i→∞
α α
where
Zt
Fi (t) = fi (λ)dλ. (4)
α
shown that a spline based approach, in general, is superior. Applications of a spline based integral
approximation are detailed in Section 6 and concluding comments are detailed in Section 7.
dk
In terms of notation f (k) (t) = dtk f (t) is used and all derivatives in the paper are with respect to
the variable t.
Mathematica has been used to generate all numerical results, graphic display of results and, where
appropriate, analytical results.
Zt n−1 h
X i
I (t) = f (λ)dλ = ck αk+1 f (k) (α) − tk+1 f (k) (t) + Rn (α, t), (5)
α k =0
where
Zt
(−1)k+1 (−1)n
ck = , Rn (α, t) = · λn f (n) (λ)dλ. (6)
(k + 1)! n!
α
For notational simplicity, it is useful to use the latter form of the definition with the former form being
implicit for the case of t = α. A Taylor series enables a function f : R → R , assumed to be (n + 1)th
order differentiable, to be written as
Zt
( t − λ ) n f (n+1) ( λ )
Rn (α, t) = dλ. (9)
n!
α
See, for example, [8] or [16] for a proof. A sufficient condition for convergence of a Taylor series is for
(n)
fmax (t − α)n
lim = 0, (10)
n→∞ n!
Math. Comput. Appl. 2018, 23, x FOR PEER REVIEW 4 of 42
( )
( − )
lim = 0, (10)
→ !
Math. ( ) Appl. 2019,(24,
Comput.
where = sup{ ) (35): ∈ [α, β]}. 4 of 40
It then follows that a nth order Taylor series for the integral
(n)
n o
where fmax = sup f (n) (t) : t ∈ [α, β] .
( ) =for the
It then follows that a nth order Taylor series (λ)integral
λ (11)
Zt
based on the point , is
I (t) = f (λ)dλ (11)
α ( )
( )= ( − α) (α) + (α, ), (12)
based on the point α, is
n−1
where X
I (t) = ck (t − α)k+1 f (k) (α) + Rn (α, t), (12)
k =0 ( )
1 ( − λ) (λ)
= , (α, ) = λ. (13)
where ( + 1)! !
t
( t − λ ) n f (n) ( λ )
Z
1
ck = , Rn (α, t) = dλ. (13)
(k + 1)! n!
3. Dual Taylor Series α
A natural
3. Dual generalization of a Taylor series is to use two Taylor series, based at different points,
Taylor Series
and to combine them by using appropriate weighting, or demarcation, functions. The result is a dual
A natural
Taylor series. generalization of a Taylor series is to use two Taylor series, based at different points,
and to combine them by using appropriate weighting, or demarcation, functions. The result is a dual
Taylor series.
3.1. Demarcation Functions
For the normalized
3.1. Demarcation Functions case of an interval [0, 1], a dual Taylor series requires two demarcations
functions, denoted and , which have the monotonic decreasing/increasing form illustrated in
For the normalized case of an interval [0, 1], a dual Taylor series requires two demarcations
Figure 1. The ideal, and normalized, demarcation function are defined according to
functions, denoted mN and qN , which have the monotonic decreasing/increasing form illustrated in
Figure 1. The ideal, and normalized, demarcation
1 function are defined according 1 to
⎧1 0< < ⎧0 0< <
⎪ 2 ⎪ 2 1
( )= 0.51 0=< 0.5
t < 12 ( )= (1 − ) = 0.5 0
0<
= t< 2
0.5 (14)
mN (t) =⎨ ⎨
0.5 1 t = 0.5 q ( t ) = m ( 1 − t ) = 1
0.5 t = 0.5 (14)
⎪ N N ⎪1
0 < <1 < <1
⎩ 0 2 21 < t < 1 ⎩ 1 2 12 < t < 1
1 mN t
qNt
t
1
Figure 1. Illustration of the normalized demarcation functions mN (t) and qN (t).
Figure 1. Illustration of the normalized demarcation functions ( ) and ( ).
For a dual Taylor series, the further requirement is that the demarcation functions do not affect
the value of the derivatives of the series at the end points of the interval being considered. This can
be achieved by the further constraints of the right and left hand derivatives, of all orders, being zero,
t
1
Figure 1. Illustration of the normalized demarcation functions m N ( t ) and q N ( t ) .
mN ( t ) k D = 10 qN ( t ) = mN ( 1 – t )
1.0
kD = 1
kD = 1
0.8
kD = 2 kD = 2
0.6
kD = 5 kD = 5
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 t
Figureof2.theGraph
Figure 2. Graph of the normalized
normalized demarcation demarcation N (t) and qm
functions mfunctions (t()t )asand
NN q N ( t )byas(16). Such
defined
functions are infinitely differentiable on (0, 1), and have right and left hand derivatives,
defined by (16). Such functions are infinitely differentiable on ( 0, 1 ) , and have of all orders,
right and left hand derivatives, of all orders, that are zero, respectively, at the
that are zero, respectively, at the points of zero and one.
points of zero and one.
For the denormalized case, and for the interval [α, β], the demarcation functions are defined
128 For the denormalized case, and for the interval [ α, β ] , the demarcation functions are defined according to
according to ! ! !
t−α t−α β−t
m(t) = mN ,t–α q(t) = qN = t – Nα
m . β–t (17)
129 m ( tβ = αm N -------------
) − q ( tβ) −=αq N -------------β =− α
m N ------------- (17)
β – α β – α β – α
Polynomial Based Demarcation Function
Math. Comput. Appl. 2019. FOR PEER REVIEW 4
Polynomial demarcation functions are of interest because, if they are associated with a Taylor
series, the resulting composite function has a known antiderivative form.
A normalized polynomial based demarcation function, of order n, and for the interval [0, 1], is the
(2n + 1)th order polynomial
" #
n
n+1 P (n+k )! k
mN,n (t) = (1 − t) · 1+ n!·k! ·t
k =1 (18)
n
n+1 P (n+k )! k
= (1 − t) · n!·k! ·t , t , 0.
k =0
For notational simplicity, the latter form is used with the former form being implicit for the case of
t = 0. This function satisfies the constraints
(k )
mN,n (0) = 1, mN,n (1) = 0, mN,n (t)t∈{0,1} = 0, k ∈ {1, . . . , n}, (19)
Math. Comput. Appl. 2019, 24, 35 6 of 40
and is antisymmetric around the point (1/2, 1/2). The associated quadrature polynomial demarcation
function is
n
X (n + k )!
qN,n (t) = 1 − mN,n (t) = mN,n (1 − t) = tn+1 · ·(1 − t)k . (20)
n!·k!
k =0
To derive (18), a useful approach is to solve for the coefficients of a (2n + 1)th order polynomial,
subject to the constraints specified by (19), and starting with the case of n = 1, 2, . . .. The coefficient
(n+k )!
form of n! · k! in (18) can be inferred from the results specified in Pascal’s triangle.
An alternative form for mN,n (t) is
+1
nP n
P (−1)u ·(n+ν)! u+ν
mN,n (t) = 1 + (n + 1) (n+1−u)!·u!·ν!
·t
u=0 ν=0,u+ν,0
+1 P
nP n
(21)
(−1)u ·(n+ν)! u+ν
= (n + 1) (n+1−u)!·u!·ν!
·t , t , 0,
u=0 ν=0
which arises from using the binomial formula on (1 − t)n+1 . For notational simplicity, the latter form is
used with the former form being implicit for the case of t = 0.
The graphs of the polynomial based demarcation functions are shown in Figure 3. For the
denormalized case, and for the interval [α, β], the demarcation functions are defined according to
! ! !
t−α t−α β−t
mn (t) = mN,n , qn (t) = qN,n = mN,n . (22)
β−α β−α β−α
Explicit forms for the demarcation functions, based on the form specified in (18), are:
" #
h β−t in+1 n n
(n+k)! t−α k
P h i h β−t in+1 P (n+k)! t−α k
h i
mn (t) = β−α 1+ n!·k! · β−α = β−α n!·k! · β−α ,
k =1 k =0
in+1
"
n
#
n (23)
(n+k)! β−t k t−α n+1 P (n+k)! β−t k
h h i h i h i
t−α P
q(t) = β−α 1+ n!·k! · β−α = β−α n!·k! · β−α .
k =1 k =0
The second form in these equations are valid, respectively, for t , α and t , β, and for notational
simplicity, are used.
m N, n ( t ) q N, n ( t )
1.0
0.8
n = 6
n = 1
0.6
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 t
Figure 3. Graph Figure
of the 3. Graph polynomial
normalized of the normalized
demarcation polynomial demarcation
functions, functions,
of orders one of the
to six, for
interval [0, 1]. orders one to six, for the interval [ 0, 1 ] .
2 (2) n (n)
(1) (t – α) f (α) (t – α) f (α)
f n ( α, β, t ) = m ( t ) f ( α ) + ( t – α )f ( α ) + ------------------------------------ + … + ------------------------------------ +
2 n!
158 (2
2 (2) n (n)
( β ) + (----------------------------------
(1) t – β) f (β) ( t – β ) f ( β )-
q ( t ) f ( β ) + ( t – β )f - + … + ---------------------------------- t ∈ [ α, β ]
Math. Comput. Appl. 2019, 24, 35 7 of 40
where m and q are the demarcation functions defined by (17). Using the Taylor series notation as
specified in (7):
fn (α, β,t) = m(t) fn (α,t) + q(t) fn (β,t). (25)
By construction:
For the case of polynomial demarcation functions, m(t) = mn (t) and q(t) = qn (t), as specified by
(23), and an explicit expression for fn (α, β, t) is
Zt Zβ
( t − λ ) n f (n+1) ( λ ) ( t − λ ) n f (n+1) ( λ )
Rn (α, β, t) = m(t) dλ − q(t) dλ, t ∈ [α, β]. (29)
n! n!
α t
3.2.1. Convergence
(n)
With the definition of fmax = sup{ f (n) (t) : t ∈ [α, β]}, it follows that a bound on the remainder
function is
(n+1) h
fmax
· m (t ) · (t − α )n+1 + q (t ) · (β − t )n+1 .
i
|Rn (α, β, t)| ≤ (30)
(n + 1)!
It then follows, from the nature of the demarcation functions, that a sufficient condition for the
convergence of a dual Taylor series for the interval [α, β] is for
(n)
fmax (β − α)n (n)
lim = 0, fmax = sup{ f (n) (t) : t ∈ [α, β]}. (31)
n→∞ n!
(n) n
f max ( β – α ) (n) (n)
176 lim ------------------------------- = 0 f max = sup { f ( t ): t ∈ [ α, β ] } (31)
n→∞ n!
Dual183 Dual
Taylor Taylor
series series approximations
approximations to thistofunction,
this function, 4, 6,4,8,6,10,
of orders
of orders 8, 10
for, for
thethe interval[0,[ 01]
interval , 1 ]and
and defined by (27),
184 are shown in Figure 4 using the polynomial demarcation functions specified by
defined by (27), are shown in Figure 4 using the polynomial demarcation functions specified by (23).(23).
1.0
f(t)
0.8
n = 8
0.6
n = 6
0.4
n = 4 n = 10
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 t
Figureof4.the 4th,
Figure 4. Graph Graph6th,of8th
theand
4th10th
, 6th , 8th
order dual and 10thseries
Taylor orderapproximations,
dual Taylor series
based on the
interval [0,approximations, based
1], for the function on the
defined byinterval
(33). [ 0, 1 ] , for the function defined by (33).
and
n
(1)
cn,k (k + 1)(t − α)k f (k) (α)+
P
Rn (α,t) = cn,0 [ f (t) − f (α)] −
Math. Comput. Appl. 2019. FOR
k=1 PEER REVIEW (36) 7
n
k +1
(t − α)k f (k) (t)[(k + 1)c n+1
( t − α ) n+1 f (n+1) ( t )
P
(−1) n,k − cn,k−1 ] + cn,n (−1)
k =1
4. Spline Approximation
A nth order spline approximation, fn , on the interval [α, β], to a function f , which is differentiable
up to order n, is a (2n + 1)th order polynomial that equals the function, in terms of value and derivatives
Math. Comput. Appl. 2019, 24, 35 9 of 40
up to order n, at the end points of the interval. A nth order spline function for the interval [α, β], thus,
has the form
fn (α, β, t) = c0 + c1 t + c2 t2 + · · · + c2n+1 t2n+1 , t ∈ [α, β], (37)
where the coefficients c0 , . . . , c2n+1 are such that the function and the spline approximation, as well as
their derivatives of order 1, . . . , n, take on the same values at the end points of the interval, i.e.,
(k )
f (k) (t) = fn (α, β, t)t∈{α,β} , k ∈ {0, 1, · · · , n}. (38)
(t−β)n+1 (t−α)n+1
where hn (t) = and gn (t) = , allows the sequential solving of the unknown coefficients
(β−α)n+1 (β−α)n+1
and leads to the result (see Appendix C):
n n−k
(β−t)n+1 (t−α)k (n+i)! (t−α)i
· f (k ) ( α ) ·
P P
fn (α, β, t) = · · +
(β−α)n+1 k =0
k!
i=0
i!·n! (β−α)i
n n−k
(40)
(t−α)n+1 (−1)k (β−t)k (n+i)! (β−t)i
f (k ) ( β ) ·
P P
· · ·
(β−α)n+1 k =0
k!
i=0
i!·n! (β−α)i
This expression can be written in the following manner which is similar in form to that of a dual
Taylor series:
n
(t−α)k
· f (k) (α) · [1 + cn,k (t)]+
P
fn (α, β, t) = mn (t) · k!
k =0
n (41)
P (t−β)k (k) (β) · [1 + d (t)]
qn ( t ) · k! · f n,k
k =0
where mn (t) and qn (t) are the denormalized polynomial demarcation functions defined by (23) and the
coefficient functions are defined according to cn,0 (t) = dn,0 (t) = 0 and
" # " #
n n
P (n+i)! (t−α)i P (n+i)! (β−t)i
− i!·n! · − i!·n! ·
i=n−k+1 (β−α)i i=n−k+1 (β−α)i
cn,k (t) = n
, dn,k (t) = n
, k ∈ {1, · · · , n} (42)
P (n+i)! (t−α)i P (n+i)! (β−t)i
i!·n! · i!·n! ·
i=0 (β−α)i i=0 (β−α)i
1 + c n, k ( t )
t = 1⁄4
1
100 300
1000
0.100
0.010
0.001 10
(β −t)2
β (–1)t- (t−α------------
------------ )t 2– α
-
219 f1 (α, β, t) = · f (α) 1 + fβ−α 0 ( α
2(t− α)
, β
+, (tt)− = α ) · f ( α⋅)f (+α ) + · f (⋅β f (
) β
1 )
+
2(β−t)
+ ( t − β ) · f (1) ( β ) (44) (43)
(β − α)2 β–α (β − αβ) – α
2 β−α
Spline Approximation
Consider the function defined by (33). Spline series approximations to this function, of orders
4, 6, 8, 10, for the interval [0, 1] and as defined by (40), are shown in Figure 6 with the 10th order
approximation visually coinciding with the function. A comparison of Figure 6 with Figure 4 shows
that a spline series provides, in general, a better approximation than a dual Taylor series of the same
order and with a dual Taylor series diverging more in the center of the interval of approximation.
224 Consider the function defined by (33). Spline series approximations to this function, of orders 4, 6, 8, 10 , for the
225 interval [ 0, 1 ] and as defined by (40), are shown in Figure 6 with the 10th order approximation visually coinciding
226 with the function. A comparison of Figure 6 with Figure 4 shows that a spline series provides, in general, a better
227 approximation than a dual Taylor series of the same order and with a dual Taylor series diverging more in the center
228Comput.
Math. of the interval
Appl. 2019, 24,of35approximation. 11 of 40
n = 8 n = 6
1.0 n = 10
f(t)
0.8
n = 4
0.6
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 t
Figureof6.4th,Graph
Figure 6. Graph ofand
6th, 8th 4th10th
, 6thorder
, 8thspline
and approximations,
10th order spline approximations,
based on the interval [0, 1],
based on the interval [ 0, 1 ] , for the function defined by (33).
for the function defined by (33).
4.2. Convergence
229 4.2 Convergence
Consider a spline approximation, as defined by (40) or (41). A sufficient condition for convergence,
i.e.,230
lim fn (α,Consider
β, t) = f a(t)spline
, is forapproximation, as defined by (40) or (41). A sufficient condition for convergence, i.e.
231n→∞ lim f ( α, β, t ) = f ( t ) , is for
n
n→∞
(n)
fmax (β − α)n (n)
n o
lim (=
n ) 0, nt ∈ [α, β], fmax = sup f (n) (t) : t ∈ [α, β] . (47)
n→∞ n
2 ·n! f max ( β – α ) (n) (n)
232 lim ⋅ ------------------------------- = 0 t ∈ [ α, β ], f max = sup { f ( t ): t ∈ [ α, β ] } (47)
n → ∞ n
The proof of this result is detailed in2 Appendix⋅ n! D.
233 The proof of this result is detailed in Appendix 4.
4.3. Spline Based Integral Approximation
234The spline
4.3 Spline Based Integral
approximation, asApproximation
defined by (40), leads to the integral equality
235 The spline approximation, as defined by (40), leads to the integral equality
Zt Xn
cn,k (t − α)k+1 f (k) (α) + (−1)k f (k) (t) + Rn (α, t)
h i
I (α, t) = f (λ)dλ = (48)
α k =0
Math. Comput.
and the following Appl.
spline 2019.integral
based FOR PEER REVIEW
approximation 10
Zt n
X
cn,k (t − α)k+1 f (k) (α) + (−1)k f (k) (t) .
h i
In (α, t) = fn (α, t, λ)dλ = (49)
α k =0
n−k
n + 1 X (n + i)! (k + i)!
cn,k = · (50)
k! i! (n + k + i + 2)!
i=0
A simpler expression is
n! (2n + 1 − k)!
cn,k = · (51)
(n − k)!(k + 1)! 2·(2n + 1)!
and this can be written, for k ≥ 1, in the form
k−1 k−1
1 Y i Y 1
cn,k = k + 1
· 1 − · , k ≥ 1. (52)
2 (k + 1 )! i=0 n (1 + (1 − i)/2n)
i=0
Math. Comput. Appl. 2019, 24, 35 12 of 40
Rt n
(t−α)k+1
· [ f (k) (α) + (−1)k f (k) (t)]
P
f (λ)dλ = 2k + 1 ( k + 1 ) !
α k =0
(t− α) (t−α)2 (t−α)3 (55)
= 2 · [ f (α) + f (t)] + 8 · [ f (1) (α) − f (1) (t)] + 48 · [ f (2) (α) − f (2) (t)]+
(t−α)4 (3) (α) − f (3) (t)] + · · ·
16·24 · [ f
t−α
I0 (α, t) = · [ f (α) + f (t)] (56)
2
t−α (t − α)2
I1 (α, t) = · [ f (α) + f (t)] + · [ f (1) (α) − f (1) (t)] (57)
2 12
t−α (t − α)2 (t − α)3
I2 (α, t) = · [ f (α) + f (t)] + · [ f (1) (α) − f (1) (t)] + · [ f (2) (α) + f (2) (t)] (58)
2 10 120
t−α 3(t−α)2 (t−α)3
I3 (α, t) = 2 · [ f ( α ) + f ( t )] + 28 · [ f (1) (α) − f (1) (t)] + 84 · [ f (2) (α) + f (2) (t)]+
(59)
(t−α)4 (3) (α) − f (3) (t)]
1680 · [ f
(1) 1 t − α (1)
R0 (α, t) = · [ f (t) − f (α)] − · f (t) (60)
2 2
(1) 1 (t − α) ( t − α ) 2 (2)
R1 (α, t) = · [ f (t) − f (α)] − · [2 f (1) (t) + f (1) (α)] + · f (t) (61)
2 6 12
(1) 1 (t−α) (t−α)2 (t−α)3
R2 (α, t) = 2 · [ f (t) − f (α)] − 10 · [3 f (1) (t) + 2 f (1) (α)] + 40 · [3 f (2) (t) − f (2) (α)] − 120 · f (3) ( t ) (62)
2
(1) 1 (t−α) (1) (t) + 3 f (1) (α)] + (t−α)
R3 (α, t) = 2 · [ f (t) − f (α)] − 14 · [4 f 28 · [2 f (2) (t) − f (2) (α)]−
(63)
(t−α)3 4
(3) (t) + f (3) (α)] + (t−α) · f (4) (t)
420 · [4 f 1680
Math. Comput. Appl. 2019, 24, 35 13 of 40
Zt
t−α
f (λ)dλ = · [ f (α) + f (t)] + R0 (α, t) = I0 (α, t) + R0 (α, t) (64)
2
α
and the area illustrated in Figure 7. The area is consistent with an affine approximation between the
function values at the end points of the interval and equals the zeroth order integral approximation
as defined by I0 (α, t). The difference between the function and an affine approximation between the
values of f (α) and f (t) defines a residual function r1 :
" # #
(λ − α)[ f (t) − f (α)]
r1 (λ) = f (λ) − f (α) + , λ ∈ [α , t . (65)
t−α
f(λ )
f(t)
(λ – α)[f(t ) – f(α)]
f ( α ) + ------------------------------------------------
t–α
f(α) + f(t)
f(α) r1 ( λ ) area = ( t – α ) ⋅ -------------------------
2
λ
α t
Figure 7.of Illustration
Figure 7. Illustration of thebyarea
the area defined defined
a zeroth byapproximation
order a zeroth ordertoapproximation
an integral andtothe
anresidual
integral
function r .and the residual function r 1 .
1
ratepoints α and
of change of tf and
( 1 ) the second residual function r .
(α) 2 2
(t – α) (1) t–α (t – α) (1)
area = ---------------- ⋅ f ( α ) ⋅ ----------- = ------------------ ⋅ f ( α )
r1 ( λ ) 4 2 8
2
t (t – α) (1) t–α –( t – α ) (1)
area = ---------------- 2⋅ – f ( t ) ⋅ ----------- = ---------------------- ⋅ f ( t )
t–α ( t –4 α ) (1) 2(1) 8
f ( λ ) dλ = ----------
2
- ⋅ [ f ( α ) + f ( t ) ] + ------------------ ⋅ [ f ( α ) – f ( t ) ] +
(1)
rate 10 of change of f ( t )
282 α (67)
3 λ
α (t – α) (2) (2)
------------------ ⋅ [ f (t α ) + f ( t ) ] + R 2 ( α, t )
120
r2 ( λ )
283 and the area illustrated in Figure 9. The third term in this expression approximates the area under a quadratic approx-
Figure 8. Illustration of the residual function (r21), the areas as ( 2 )defined by linear change at the points α
284 Figure
imation8.to the
Illustration
residual of the residual
function r 2 , based
and t and the second residual function r2 .
function
on f r(1α, )the
andareas
f (as
t ) ,defined
and withbya linear change
different at the term of 120 ver-
denominator
sus 48α . and
285 points t and the
For higher second
order residual function
approximations the term r 2 .of 120 approaches 48 .
r2 ( λ )
t 2
t–α (t – α) ( 1λ
) (1)
f ( λ ) dλα = ----------
2
t
- ⋅ [ f ( α ) + f ( t ) ] + ------------------
10
⋅ [ f ( α ) – f ( t( t–) ]α+) 3 ( 2 )
area = ------------------ ⋅ f ( t )
282 α 48 (67)
3
3 (2)
λ
α t
r2 ( λ )
Figure 8. Illustration of the residual function r 1 , the areas as defined by linear change at the
Math. Comput. Appl. 2019, 24, 35 14 of 40
points α and t and the second residual function r 2 .
λ
α t (t – α)
3
(2)
area = ------------------ ⋅ f ( t )
48
3
(t – α) (2)
2 area = ------------------ ⋅ f ( α )
(λ – α) (2) 2 48
(λ – t)
-------------------- ⋅ f ( α ) (2)
2 ------------------ ⋅ f ( t )
2
Figure 9. Illustration of the areas defined by a quadraticbyapproximation
Figure 9. Illustration of the areas defined a quadratic approximation
to the residualtofunction
the residual
r and 2
(2) (2)
based onfunction r 2 and
f (2) (α) and (t). on f
f (2) based ( α ) and f (t) .
286 For the convergent case, the approximation to the integral, as specified by defined by (49), converges to the form
For the convergent case, the approximation to the integral, as specified by defined by (49),
287 specified by (55) as n → ∞ and this form is consistent with the successive area approximations illustrated in
288 converges to the form
Figure 7, Figure Figure 9. by (55) as n → ∞ and this form is consistent with the successive area
8 and specified
approximations illustrated in Figures 7–9.
289 4.5 Determining Region of Integration for a Set Error Bound
4.5. Determining Region of Integration for a Set Error Bound
290 For a positive function, a bound on the region of integration, for a set error level, can be determined by consider-
291 ing For a positive
the area function,
defined a bound
by the first ontime
passage the region of integration,
of the function for a set
to a specified errorConsider
level. level, can be determined
a positive function, which
292 byhas
considering the area
a first passage defined
time to by rtheatfirst
the level B t =passage
t B , as time of the
illustrated in function
Figure 10.toThe
a specified
followinglevel. Consider
integral bound a
holds
positive function, which has a first passage time to the level rB at t = tB , as illustrated in Figure 10.
The following integral bound holds
ZtB
Math. Comput. Appl. 2019. FOR PEER REVIEW 13
f (λ)dλ ≤ rB tB = εB , (68)
0
where
tB = inf t : | f (t)| = rB
tB (69)
= inf t : t f (t)| = tB rB = εB
293 f ( λ ) dλ ≤ r B t B = εB
Thus, for a bound εB on the integral of a positive function f , a bound on the interval of integration is
(68)
0
294 where
tB = inf t : t f (t)| = εB , (70)
t B = inf { t: f ( t ) = r B }
295first passage time of t f (t) to εB .
i.e., the (69)
= inf { t: t f ( t ) = t B r B = ε B }
f(t)
rB tB
area = ε B = bound on f ( λ ) dλ
0
t
tB
Figure 10. The area bound defined by the first passage time of a positive function to a set
Figure 10. The area bound defined by the first passage time of a positive function to a set level.
level.
296 Thus, for a bound, ε B on the integral of a positive function f , a bound on the interval of integration is
Zt Zt
(1)
I (α, t) = f (λ)dλ = In (α, t) + Rn (α, t), Rn (α, t) = Rn (α, λ)dλ, (71)
α α
(1) (1)
where Rn (α, t) is specified by (53). As Rn (α, t) is known, it follows that a bound on the region of
integration can be specified according to
(1)
tB = inf{t : t|Rn (t)| = εB } (72)
Example
R tB sin(t)
Consider determining the region of integration, tB , for the integral 0 1+t dt and for an error
sin(t)
bound of εB = 10−3
using the integral approximation specified by (49). The graph of f (t) = 1+t is
shown in Figure 11 and results are tabulated in Table 1. The specified region of integration, as expected,
is conservative.
Table 1. Interval of integration for a spline based approximation and for a specified error bound of
εB = 10−3 .
sin ( t )-
f ( t ) = -------------
1+t
0.4
0.3
0.2
0.1
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 t
sin(t)
sinf (t()t )= 1+t over the interval [0,
Figure Graph11.ofGraph
11. Figure f ( t ) =of-------------
- over the interval [ 0, π ] . π ].
1+t
5. Summary and Comparison of Integral Approximations
Table 1. Interval of integration for a spline based approximation and for a specified error
The function and integral approximations detailed
–3 above are summarized in Table 2. The remainder
bound of ε B = 10 .
terms associated with the integral approximations are summarized in Table 3.
Magnitude of error
Order of integral in integral approx.
approx: n tB over [ 0, t B ]
3 1.17 –4
1.5 ⋅ 10
4 1.47 –4
1.3 ⋅ 10
6 1.96 –4
1.0 ⋅ 10
Math. Comput. Appl. 2019, 24, 35 16 of 40
m
−(t − ti )2 2 t2
, σ = FWHMi ,
X
f (t) = ai exp i (73)
i=1
2σ2i 8 ln(2)
where m is an outcome of a Poisson random variable with parameter λ = 4 (the zero case excluded),
ai , i ∈ {1, . . . , m}, are independent outcomes of random variables with a normal distribution with
zero mean and unit variance, ti , i ∈ {1, . . . , m}, are independent outcomes of random variables with a
uniform distribution on the interval [0, 1] and t2FWHM , i ∈ {1, . . . , m}, are independent outcomes of
i
random variables with a uniform distribution on [0, 2]. Examples of signals are shown in Figure 12.
R1
For 1000 independently generated signals, the proportion of approximations to 0 f (λ)dλ, with a
relative error of less than 0.01, is detailed in Table 4 for the four integral approximations. The results
show the clear superiority of the spline based integral approximation and simulation results for other
types of signals indicate that this holds more generally.
1
322 approximations to f ( λ ) dλ , with a relative error of less than 0.01 , is detailed in Table 4 for the four integral approx-
0
323 imations. The results show the clear superiority of the spline based integral approximation and simulation results for
324Comput.
Math. other types
Appl. of24,
2019, signals
35 indicate that this holds more generally. 17 of 40
Table
Table 4. Proportion 4. Proportion
of integral of integral
approximations, approximations,
based based
on a set of 1000 test on a set of with
functions, 1000atest functions, with a relative
relative
error less than 0.01.
error less than 0.01.
-1
-2
-3
-4
6.1.
334Exponential Function Function
6.1 Exponential Approximation
Approximation
335The spline
The based integral
spline based series,
integral of order
series, n, leads
of order to to
n ,leads thethe
following
followingseries
series approximation forthe
approximation for the
exponential func-
exponential
336 tionfunction
t t2 t3 t4
1+ 2 + 22 ·2!
+ 23 ·3!
+ 24 ·4!
+···
exp(t) ≈ (75)
t t2 t3 t4
1− 2 + 22 ·2!
− 23 ·3!
+ 24 ·4!
−···
The series converges for all t ∈ R. The proof of this result is detailed in Appendix F.
2 2 2 ⋅ 2! 2 3 ⋅ 3! 2 4 ⋅ 4!
339 exp ( t ) = ------------------------------------------------------------------------------------------ (75)
2 3 4
t t t t
1 – --- + --------------- – --------------- + --------------- – …
2 2 2 ⋅ 2! 2 3 ⋅ 3! 2 4 ⋅ 4!
340 TheMath.
series converges
Comput. ∈ R . The proof of this result is detailed in Appendix 6.
for all24,t 35
Appl. 2019, 18 of 40
0.010
8th order
0.001
10-4
10-5
10-6
0 2 4 6 8 10 t
High precision results can be obtained as m is increased. Results that are indicative of the improvement,
with increasing levels of fractional power, are detailed in Table 5 where the case of approximating e
is considered.
Math. Comput. Appl. 2019, 24, 35 19 of 40
n
(−1)k
2(t − 1) X
k + 1 k
ln(t) ≈ · 1 +
cn,k (−1) (k − 1)!(t − 1) 1 +
, (80)
t+1
k =1
t k
where cn,k is specified by (51). The proof of this result is detailed in Appendix G.
A Taylor series, based on the point t = 1, for the natural logarithm is
n
X (−1)k+1 (t − 1)k
ln(t) = . (81)
k
k =1
The relative error in this series, and the spline based series defined by (80), is detailed in Figure 14.
66 The relative error in this series,
Simulation resultsand the spline
indicate based series
a region defined by close
of convergence (80), is 0.15 < tin
todetailed < Figure 14.spline
6 for the Simulation
basedresults
series-proof
67 indicate a region
of aof convergence
definitive boundclose
is to
an 0.15 < t < 6problem.
unsolved for the spline based series - proof of a definitive bound is an
68 unsolved problem.
Taylor series
1
10-4
10-5
10-6
0.2 0.5 1 2 5 t
69 6.2.1 Example
70 A fourth order approximation is
2 2 2 3 3 4 4
2(t – 1) (t – 1) (t + 1 )(t – 1) ( t – 1)(t – 1) (t + 1)( t – 1 )
71 ln ( t ) ≈ ------------------ 1 + ------------------ – ------------------------------------- + ------------------------------------ – ------------------------------------- (82)
Math. Comput. Appl. 2019, 24, 35 20 of 40
6.2.1. Example
A fourth order approximation is
(t − 1)2 (t2 + 1)(t − 1)2 (t3 − 1)(t − 1)3 (t4 + 1)(t − 1)4
2(t − 1)
ln(t) ≈ 1 + − + − . (82)
t+1 9t 72t2 504t3 5040t4
it follows that
t1/m t1/m
t1/m
Z Z
1/m m h 1/m i m
ln(λ)dλ = ln(t) − t + 1 ⇒ ln(t) = 1/m t − 1 + 1/m ln(λ)dλ. (84)
m t t
1 1
Using the spline series approximation specified in (80), it follows that a nth order series approximation
for ln(t), with higher precision and a greater range of convergence, is
n
(−1)k
2mt1/m 1 2m
k +1
X
k +1 1/m
ln(t) ≈ · 1 − 1/m + · cn,k (−1) (k − 1)!(t − 1) 1 + k/m (85)
1 + t1/m t 1 + t1/m k=1 t
where cn,k is specified by (51). Results that are indicative of the improvement in precision, with
increasing levels of fractional power, are detailed in Table 6.
Table 6. Relative error in evaluation of ln(t) for the case of t = 100. For the case of m = 1 the relative
error is high for all orders of approximation.
Z1
1
ln(2) = dt = 0.6931471805 · · · . (86)
1+t
0
Math. Comput. Appl. 2019, 24, 35 21 of 40
whilst Taylor series for ln(x), based on the points at 1 and 2, yield
∞ ∞
X (−1)k−1 X 1
ln(2) = , ln(2) = , (88)
k =1
k
k =1
k · 2k
i.e.,
1 1 1 1 1 1
ln(2) = 1 − + − +··· , ln(2) = + + +··· . (89)
2 3 4 2 2·4 3·8
The second series arises by finding an approximation for ln(1) based on the Taylor series at the point 2.
The relative errors in the Taylor series for orders 2, 4, 8, 16, respectively, are: 0.28, 0.16, 0.085 and
0.044 for the first series and 0.098, 0.016, 5.7 × 10−4 and 1.2 × 10−6 for the second series. The relative
errors in the series defined by (87), for orders 2, 4, 8, 16, are: 0.16, 0.085, 0.044 and 0.022.
A second order series specified by (85) is
2 2
21/m − 1 (21/m − 1) (21/m − 1) (22/m + 1)
ln(2) ≈ 2m · 1/m 1 + − (90)
2 + 1 10 · 21/m 120 · 22/m
and yields an approximation with relative errors, respectively, of 1.3 × 10−4 , 1.3 × 10−10 , 1.3 × 10−16
and 1.3 × 10−22 for the cases of m = 1, 10, 100, 1000.
Zx0
1 x
I0 = dt = t1/h | h0 = x1/h
o − z0 ⇒ xo1/h = z0 + I0 . (91)
ht1−1/h z0
zh0
x1/h
o ≈ z1 = z0 + approximation to I0 . (92)
Replacing zh0 by zh1 in (91) is the first step in an iterative algorithm to establish an accurate approximation
to x1/h
o . The requirement for such an algorithm is a suitable approximation for the integral defined by
I0 and a nth order spline integral approximation is useful.
n k
" # (94)
P h k +1 Q 1
1
νi + 1 = − cn,k (x0 − zi ) · j − h · k +1
k =0 j=0 x0
Here cn,k is specified by (51). The proof of this algorithm is detailed in Appendix H.
1 2 3(h−1) 3 (h−1)(2h−1)
ui+1 = (x0 − zhi ) · − (x0 − zhi ) · + (x0 − zhi ) · −
2hzh−1
i
28h2 z2h−1
i
84h3 z3h−1
4 (h−1)(2h−1)(3h−1)
i (95)
(x0 − zhi ) ·
1680h4 z4h−1
i
Indicative results for the efficacy of the iterative algorithm for evaluation of a rational number to a
fractional power, based on third and sixth order spline approximations, are detailed in Table 7.
g (x)
q i
1 + g2 ( x )
Z i
1
θ = 2i · p dλ, i ∈ {1, 2, . . .}. (97)
0
1 − λ2
This result is proved in Appendix I. To determine an approximation for θ, and, hence, acos(x), asin( y)
and atan(y/x), first, gi (x) needs to be specified. Second, an approximation to the integral needs to be
specified and the spline based integral approximation is efficacious.
Math. Comput. Appl. 2019, 24, 35 23 of 40
Here:
gi (x) = tan(θ/2i ), si = sin(θ/2i ), ci = cos(θ/2i ). (99)
p(0, t) = 1,
d (104)
p(k, t) = 1 − t2 dt p(k − 1, t) + (2k − 1)tp(k, t)
As an example
p(6, t) = 45 5 + 90t2 + 120t4 + 16t6 . (105)
These results arise from the spline based integral approximation as specified by (49) and (51) and by
noting that
1 p(k, t)
f (t) = 1/2
⇒ f (k ) ( t ) = , k ∈ {0, 1, . . .}, (106)
2
[1 − t ] [1 − t2 ](2k+1)/2
Math. Comput. Appl. 2019, 24, 35 24 of 40
where the algorithm for determining the numerator polynomial p(k,t) is specified by (104). Substitution
of this result into (49) yields
k+1 k
g ( x )
( −1) pk, q i
n
1+ g2i (x)
i
X gi ( x )
θ≈2 cn,k q p[k, 0] +
.
(107)
2
g 2 (x,y) ( 2k + 1 ) /2
1 + gi (x)
k =0 1− i
1+ g2i (x,y)
where, √for the case of i = 4, g4 (x) is specified in (102). The use of g4 (x) in (108), for the case of
x = 1/ 2, yields an approximation to atan(1) = π/4 and, hence, π with a relative error of 1.5 × 10−14 .
Weisstein [19] provides a good overview of approaches for calculating π. Further, and indicative,
results are tabulated in Table 8. High precision results can be established, for relatively low order
spline based integral approximations, by using a high order of angle subdivision.
6.5. Series for Integral Defined Functions: The Fresnel Sine Integral
The spline based integral approximation, as specified by (49), can be used to define series
approximations for integral defined functions. As an example, consider the Fresnel sine integral
Rt
0
sin λ2 dλ. To establish a spline based approximation for this integral the kth derivative of
Math. Comput. Appl. 2019, 24, 35 25 of 40
f (t) = sin t2 is required. This can be specified by using the quadrature signal fq (t) = cos t2
and is
f (k) (t) = pk (t) f (t) + qk (t) fq (t), k ∈ {0, 1, . . .}, (109)
where
p0 (t) = 1, q0 (t) = 0,
p1 (t) = 0, q1 (t) = 2t, (110)
474 It then follows that a spline based
(1) approximation to the Fresnel(1sine
) integral is:
pk (t) = pk−1 (t) − 2tqk−1 (t), qk (t) = qk−1 (t) + 2tpk−1 (t), k ≥ 2
t n
It then follows that a spline based approximation to the Fresnel sine integral is:
2 k+1
tsin ( λ ) d λ ≈ c n, k t q k ( 0 ) +
n
0R k = 2, 6, P
10 ,…
475 sin(λ2 )dλ ≈ cn,k tk+1 qk (0)+ (111)
n
k=2,6,10,... n
0
2 k k+1 2 k k+1 (111)
n n
sin (sin
t )(t2 ) c n,ck ( –(1−1
P ) t)k tk+p1kp( t()t+
) cos
+ ( t ()t2 ) P
cos c nc, k ((–−1
1 ))kt tk+1qqk (( tt))
n,k k n,k k
k = 0k=0 k =k=00
t Rt
3 5t3 7 t7 409t55 997t9 9 t13 13 31t3 3 7 − 41t117 11
h i h i
2 2 t 2 5651t
2sin(λ )dλ
5t ≈ 156 −t 144144 + sin2(t )t 2 − 8580 + 1081080
409t t + cos(t ) −2156 +31t
997t − 270270 720720 5651t 41t(112)
sin (0λ
540540
477 ) dλ ≈ --------- – ------------------ + sin ( t ) --- – ------------- + --------------------- – ------------------ + cos ( t ) – ---------- + ------------------ – ------------------ (112)
156 144144 2 8580 1081080 270270 156 720720 540540
0
The error in a sixth order approximation to the integral is illustrated in Figure 15. The relative error is
478 The error in aless
significantly sixth order
than forapproximation to theseries
a standard Taylor integral is illustrated indefined,
approximation Figure 15. The
e.g., relativeaccording
[20,21], error is significantly
to
479 less than for a standard Taylor series approximation defined, e.g. [20], [21], according to
Zt t ∞
22
∞
X (−1)kk t4k 4k+3
+3 t33 t77 t11
11
sin(λ )dλ ≈ ( –1 ) t = t −t +t −.... (113)
480
0
sin ( λ ) dλ ≈ k =0 (----------------------------------------------
4k + 3) · (2k + 1)-! = ---
( 4k + 3 ) ⋅ ( 2k + 1 )!
3- – -----
42
3 42 1320
-–…
- + -----------
1320 (113)
0 k=0
481 For
For integration overthethe
integration over interval [ 0, [0,
interval 1.5 ]1.5],
, thethe
truetrue integral
integral is 0.778238
is 0.778238… . . .a and
and a 6th spline
6th order order approximation
spline
approximation yields a relative error–5 of −1.4 × 10−5 .
482 yields a relative error of – 1.4 × 10 .
spline series approx.
1.0
0.6
0.4
0.2
Taylor series approx.
0.0
-0.2
0.0 0.5 1.0 1.5 2.0 2.5 3.0 t
2
Figure 15. Graph of approximations (order 6) to the integral of sin(t2 ).
Figure 15. Graph of approximations (order 6) to the integral of sin ( t ) .
6.6. Definite Integrals
t n k i k
sin ( λ ) ( – 1 ) k! ( k – i )π ( –1 )
- ⋅ sin ------------------ + sin t + (------------------
k – i )π ------------------------
k+1
488 ---------------
1+λ
- dλ ≈ c n, k t ⋅ ----------------
( k – i )! 2 2
⋅
i+1
(114)
0 k=0 i=0
(1 + t)
489 arising from (49), with c n, k as specified by (51), and based on the result
Math. Comput. Appl. 2019, 24, 35 26 of 40
6.6.1. Example 1
Consider the approximation
arising from (49), with cn,k as specified by (51), and based on the result
k
dk X k!
k
[ u ( t ) ν ( t )] = · u(k−i) ν(i) (t), (115)
dt (k − i)!i!
i=0
(−1)i i!
h i
where u(t) = sin(t), ν(t) = 1
1+t , u(i) (t) = sin t + iπ
2 and ν(i) (t) = .
(1 +t )i+1
From (114), the following definite integral approximation
Zπ n
k
X (−1)i k!
" # k +1
sin(λ) X ( k − i ) π ( −1 )
cn,k πk+1 ·
dλ ≈ · sin 1 +
(116)
1+λ (k − i)! 2 i + 1
( 1 + π )
0 k =0 i=0
where
p(ν) (t)
dn,ν = ν! ,
t→0
n
"
k
# (118)
(−1)i k!
P k−1
P (k−i)π n k +1 n−i−1
p(t) = cn,k t · (k−i)!
· sin 2 · [(1 + t) + (−1) (1 + t) ]
k =0 i=0
Zπ
3 33π 4865π2 11105π3 5669π4 145651π5 2257π6
sin(λ) π2 13 + 26 + 1716 + 3432 + 2970 + 308880 − 332640 −
dλ ≈ · 29977π7 157π8 1333π9 101π10 101π11
(119)
1+λ (1 + π)6 2882880 + 154440 − 4324320 + 864864 − 2882880
0
The true value of the integral is 0.84381081 . . . and the relative errors in the series approximation for
n = 4, 8, 16, 64, 128, 256, respectively, are 0.035, 3.3 × 10−3 , 3.9 × 10−5 , 3.2 × 10−16 , 9.2 × 10−31 and
1.1 × 10−59 .
∞
X (−1)k
G= = 0.9159655941 (120)
k =0 (2k + 1)2
π/4
Z
G=2 ln[2 cos(t)]dt. (121)
0
Math. Comput. Appl. 2019, 24, 35 27 of 40
N (k, t)
f (t) = ln[2 cos(t)], f (k ) ( t ) = , k ∈ {1, 2, . . .}. (123)
cos (t)k
Here:
N (1, t) = − sin(t),
d (124)
N (k, t) = dt N (k − 1, t) · cos(t) + (k − 1)N (k − 1, t) sin(t), t ≥ 2
A sixth order approximation is
and has a relative error of 1.7 × 10−8 . A sixth order series approximation, as specified by (120), yields
an approximation with a relative error of −2.7 × 10−3 .
Z+∆m−1
α
X t−α
I= f (λ + i∆)dλ, ∆= (126)
m
α i=0
where
m−1
X
g(t) = f (t + i∆), t ∈ [α, α + ∆]. (128)
i=0
6.7.1. Example
Consider the integral
Zt t/mm−1
Z
X sin(λ + q∆)
sin(λ)
I= dλ = dλ, ∆ = t/m, (129)
1+λ 1 + λ + q∆
0 0 q=0
which has the approximation detailed in (114) and where the demarcation of the interval [0, t] into m
sub-intervals of measure ∆ = t/m has been used. It then follows that
n
X tk+1
I≈ cn,k · k + 1
· [ g(k) (0) + (−1)k g(k) (t/m)], (130)
k =0
m
Math. Comput. Appl. 2019, 24, 35 28 of 40
where
m−1
X sin(t + q∆)
g(t) = (131)
1 + t + q∆
q=0
m−1 k
(k − i)π (−1)i
" #
(k )
X X k!
g (t) = · sin t + q∆ + ·
.
i + 1
(132)
(k − i)! 2
(1 + t + q∆)
q=0 i=0
For the case integration over the interval [0, π], the use of four subdivision intervals yields,
for the case of spline based integral approximations of orders n = 4, 8, 16, 64, 128, 256, relative errors,
respectively, with magnitudes of 1.0 × 10−6 , 4.3 × 10−11 , 9.8 × 10−20 , 4.5 × 10−71 , 2.7 × 10−139 and
1.4 × 10−275 . Such results indicate the significant improvement in convergence when compared with
the non-subdivision case as detailed in Section 6.6.1.
For the case integration over the interval [0, 100π], the use of 256 subdivision intervals yields,
for the case of spline based integral approximations of orders n = 4, 8, 16, 64, 128, 256, relative errors,
respectively, with magnitudes of 5.3 × 10−5 , 3.3 × 10−8 , 1.6 × 10−14 , 7.5 × 10−52 , 2.1 × 10−101 and
2.3 × 10−200 . The true integral is 0.618276689 . . ..
which has a free charge density of ρ f on it, generates an electric field at a point (x, y, z) away from the
ring of
Z2π [x − a · cos(t)]i + [ y − a · sin(t)] j + [z − b]k
aρ f
E(x, y, z) = · dt, (134)
4πεo [x2 + y2 + z2 + a2 − 2ax cos(t) − 2ay sin(t) − 2bz]3/2
0
where εo is the permittivity of free space. As an example, consider the x component of the electric field
which is defined by the integral
Z2π Z2π
aρ f [x − a· cos(t)] aρ f
EX (x, y, z) = · i3/2 dt = 4πε · f (t)dt, (135)
4πεo h
o
0 r2o − ao cos(t − φ) 0
where
x − a· cos(t)
f (t)= h i3/2 ,
r2o − ao cos(t − φ) (136)
q
r2o = 2 2 2
x + y + z + a − 2bz, 2
ao = 2a x2 + y2 , φ = atan( y/x).
Using (127), and demarcation of [0, 2π] into m sub-intervals of measure ∆ = 2π/m, this component
of the electric field can be approximated according to
aρ f n
X
EX (x, y, z) ≈ · cn,k ∆k+1 [ g(k) (0) + (−1)k g(k) (∆)], (137)
4πεo
k =0
where
m−1
X m−1
X
(k )
g(t) = f (t + i∆), g (t) = f (k) (t + i∆) (138)
i=0 i=0
Math. Comput. Appl. 2019, 24, 35 29 of 40
and
N (k, t)
f (k ) ( t ) h i(2k+3)/2 , k ∈ {0, 1, . . .}. (139)
r2o − ao cos(t + φ)
N ( 0, t ) = x – a ⋅ cos ( t )
( 0, t ) = x – a ⋅ cos ( t )
563 algorithm forNdetermining
The d N(k, t) is 2 ( 2k + 1 ) (140)
563 N ( k, t ) =d N ( k – 1, t ) ⋅ [2r o – a o cos ( t – φ ) ] –( 2k -------------------
+21 ) - N ( k – 1, t )a o sin ( t – φ ) (140)
d t
N ( k, t ) = N ( k – 1, t ) ⋅ [ r o – a o cos ( t – φ ) ] – -------------------- N ( k – 1, t )a o sin ( t – φ )
2
564 For example = x −d ta· cos(via
N (0, t)(simplification t), Mathematica):
564 For example (140)
N (k, t(simplification N (k −via 1, tMathematica):
h i (2k+1)
d
) = dt )· r2o − ao cos(t − φ) − 2 N (k − 1, t)ao sin(t − φ).
(4) – a cos ( t ) 3a o [ a ⋅ cos ( φ ) + 2x cos ( t – φ ) – 15a ⋅ cos ( 2t – φ ) ]
( 4f ) ( t ) = -----------------------------------------------------
– a cos ( t ) 3 ⁄
- +3a---------------------------------------------------------------------------------------------------------------------------
2 o [ a ⋅ cos ( φ ) + 2x cos ( t – φ ) – 15a5 ⋅⁄ 2cos ( 2t – φ ) ] +
f (example
For t ) = -----------------------------------------------------
2
– a o cos ( t – φ )3]⁄ 2- +via
[2r o (simplification
2
---------------------------------------------------------------------------------------------------------------------------
Mathematica): 4 [2r o – a o cos ( t – φ )5]⁄ 2 +
[ r o – a o cos ( t – φ ) ] 4 [ r o – a o cos ( t – φ ) ]
2
) (15a o [ – 2 x + 14a ⋅ cos ( t ) + 14x cos ( 2t –(φ 2φ)+) 2x – acos ⋅ cos ( t)−15a·cos
– 2φ ) – (25a ⋅ )]cos +( 3t – 2φ ) ]
−a cos(t) 3a o [a·cos (t−φ 2t−φ
f (415a t)2 [= – 2 x + 14a ⋅ cos ( t ) + 3/214x
+cos
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
( 2t – 2φ ) – a ⋅ cos ( t – 2φ )
5/2 – 25a ⋅ cos ( 3t – 2φ ) ] - +
565 o 2
[ro −ao cos(t−φ)] 2
2
4[ro −ao cos7(t−φ ⁄ 2 )] (141)
565 ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
15a2o [−2x+14a·cos(t)+16 [2rcos a o cos)(−a·cos t – φ )7(]⁄t−2φ -+
14x o –(2t−2φ 2 )−25a·cos(3t−2φ)]
+
(141)
16 [ r o –2 a o cos ( t – φ ) ]7/2 (141)
3 16[ro −ao cos(t−φ)] 2 4 4
3 o [ a 105a ⋅ cos3 ([a·cos φ ) –(6x (cos
t –(φt−φ ) +)+ ⋅ cos(2t−φ ( 2t –)]φsin ) ] sin ( t –)2 φ2) 945a [ xcos–(at)]⋅ sin cos((t−φ t ) ] )sin
105a cos 5a5a·cos 945a 4 [4x−a 4 (t – φ)
φ ) −6x ( t−φ o 4
--------------------------------------------------------------------------------------------------------------------------------------------------------
– φ ) ] sin ( t – φ ) +- +945a -----------------------------------------------------------------------------
) ] sin ( t – φ ) -
105a o [ a ⋅ cos ( φ ) – 6x cos2 ( t –2 φ ) + 5a ⋅ cos (⁄ 2t o2[ x –2 a ⋅ cos ( t 11/2
o o
π 3π
g ( t ) = f ( t ) + f t +π --- + f ( t + π ) + f t +3π------
g ( t ) = f ( t ) + f t + --- + f ( t + π ) + f t + ------2
2
2
( 10, 1, 1 ) 2
0.4 ( 10, 1, 1 )
0.4
( 1, 1, 1 )
0.2 ( 1, 1, 1 )
0.2 ( 1, 1 ⁄ 10, 1 ⁄ 10 )
0.0 ( 1, 1 ⁄ 10, 1 ⁄ 10 )
0.0
( 1 ⁄ 10, 1 ⁄ 10, 1 ⁄ 10 )
-0.2 ( 1 ⁄ 10, 1 ⁄ 10, 1 ⁄ 10 )
-0.2
-0.4
-0.4
0.0 0.2 0.4 0.6 t 0.8 1.0 1.2 1.4
0.0 0.2 0.4 0.6 t 0.8 1.0 1.2 1.4
Figure
Figure17. Graph
17. of g(t).ofFor
Graph g ( tthe casethe
) . For of case
(10, 1,of 1)( 10
a ,scaling
1, 1 ) afactor
scalingoffactor
10 hasofbeen used;
10 has forused;
been the case of case of
for the
(1,Figure ⁄ 17.
( 1, 11/10)
1/10, Graph
10, 1a ⁄scaling of g ( t )factor
10 ) a scaling
factor . For
of the0.1
0.1 of
has case ( 10used.
hasofused.
been been , 1, 1 ) a scaling factor of 10 has been used; for the case of
( 1, 1 ⁄ 10, 1 ⁄ 10 ) a scaling factor of 0.1 has been used.
Indicative results are detailed in Table 9 and the use of four sub-intervals yields, in general,
acceptable levels of error apart from the case of points close to the ring of charge where the integrand
varies rapidly.
Table 9. Magnitude of relative error in evaluation of EX (x, y, z) using four sub-intervals and eight
sub-intervals for the point (1, 0.1, 0.1).
7. Conclusions
This paper has introduced the dual Taylor series which is a natural generalization of the classic
Taylor series. Such a series, along with a spline based series, facilitates function and integral
approximation with the approximations being summarized in Table 2. In comparison with a
antiderivative series, a Taylor series and a dual Taylor series, a spline based series approximation to an
integral, in general, yields the highest accuracy for a set order of approximation. A spline based series
for an integral has many applications and indicative examples include a series for the exponential
function, which coincides with a Padé series, new series for the logarithm function as well as new
series for integral defined functions such as the Fresnel Sine integral function. Such series are more
accurate, and have larger regions of convergence, than Taylor series based approximations. The spline
based series for an integral can be used to define algorithms for highly accurate approximations for
the logarithm function, the exponential function, rational numbers to a fractional power and the
inverse sine, inverse cosine and inverse tangent functions. Such algorithms can be used, for example,
to establish highly accurate approximations for specific irrational numbers such as π and Catalan’s
constant. The use of sub-intervals allows the region of convergence for an integral to be extended.
The results presented are not exhaustive and other applications remain to be found.
Acknowledgments: The support of Prof. Zoubir, SPG, Technical University of Darmstadt, Darmstadt, Germany,
who hosted a visit where significant work on this paper was completed, is gratefully acknowledged. Prof. Zoubir
suggested the use of a set of test functions for a comparison of the integral approximations.
Conflicts of Interest: The author declares no conflict of interest.
Rn (α, β, t) = f (t) − fn (α, β, t) = [m(t) + q(t)] f (t) − [m(t) fn (α, t) + q(t) fn (β, t)]
(142)
= m(t)[ f (t) − fn (α, t)] + q(t)[ f (t) − fn (β, t)]
Using the result for the error in a nth order Taylor series approximation, as specified in (9), i.e.,
Zt Zβ
( t − λ ) n f (n+1) ( λ ) ( t − λ ) n f (n+1) ( λ )
Rn (α, t) = dλ, Rn (β, t) = − dλ, (143)
n! n!
α t
Math. Comput. Appl. 2019, 24, 35 31 of 40
Zt Zβ
( t − λ ) n f (n+1) ( λ ) ( t − λ ) n f (n+1) ( λ )
Rn (α, β, t) = m(t) dλ − q(t) dλ, t ∈ [α, β]. (144)
n! n!
α t
Rt
" #
+1 P
nP n n (λ−α)k f (k) (α)
(−1)u ·(n+ν)! λ−α u+ν
h i P
In ( t ) = (n + 1) (n+1−u)!·u!·ν!
· t−α · k! dλ+
α" u=0 ν=0 k =0
(146)
Rt
#
+1 P
nP n n (λ−t)k f (k) (t)
(−1)u ·(n+ν)! t−λ u+ν
h i P
(n + 1) (n+1−u)!·u!·ν!
· t−α · k! dλ
α u=0 ν=0 k =0
Interchanging the order of summations and integration, and evaluation of the integrals, yields the
required result:
+1 P
nP n n
(−1)u ·(n+ν)! (t−α)k+1
· [ f (k) (α) + (−1)k f (k) (t)]
P
In ( t ) = (n + 1) (n+1−u)!·u!·ν!·k!
· u+ν+k +1
u=0 ν=0 k =0 (147)
n
k +1
· [ f (k) (α) + (−1)k f (k) (t)]
P
= cn,k · (t − α)
k =0
where
n+1
n
n+1 X (−1)u X (n + ν)! 1
cn,k = · · · . (148)
k! u! · (n + 1 − u)! ν! u + ν + k + 1
u=0 ν=0
Zt Zt
Rn (α, t) = [ f (λ) − fn (α, t, λ)]dλ = f (λ)dλ − In (t), (149)
α α
which implies
(1) (1)
Rn (α, t) = f (t) − In (t). (150)
Differentiation of In (t), and use of the result cn,0 = 1/2, yields the required result:
n
(1)
cn,k (k + 1)(t − α)k f (k) (α)+
P
Rn (α, t) = cn,0 [ f (t) − f (α)] −
k =1 (151)
n
(−1)k+1 (t − α) f k (k )
(t)[cn,k (k + 1) − cn,k−1 ] + cn,n (−1)n+1 (t − α)n+1 f (n+1) (t)
P
k =1
Math. Comput. Appl. 2019, 24, 35 32 of 40
Using these results, (39) allows the unknown coefficients to be sequentially solved. For example:
etc. Sequential solving of the equations yields the result stated in (40).
To rewrite (40) in a form that is similar to the form of a dual Taylor series, both the numerator and
n n
P (n+i)! (t−α)i P (n+i)! (β−t)i
denominator expressions can be multiplied by i! · n! · i for the first term, and i! · n! · i
i=0 (β−α) i=0 (β−α)
for the second, to yield
n−k
P (n+i)! (t−α)i
i!·n! · (β−α)i
" #
(β −t)n+1 n (t−α) i n (t−α) k
(n+i)!
· f (k ) ( α ) · i=0
P P
fn (α, β, t) = · · · n +
(β − α )n+1 i=0
i!·n! (β − α)i k =0
k! P (n+i)! (t−α)i
·
i!·n! (β−α)i
i=0
n−k
P (n+i)! (β−t)i
(156)
i!·n! · (β−α)i
" #
(t−α) n+1 n (β −t) i n k
(−1) (β −t) k
(n+i)!
· f (k ) ( β ) · i=0
P P
· · · n (n+i)! (β−t)i
(β − α )n+1 i=0
i!·n! (β − α)i k =0
k! P
·
i!·n! (β−α)i
i=0
With the definition of the polynomial demarcation functions, as specified by (23), it follows that
n−k
+i)! (t−α)i
n (
P (ni!·n! ·
t−α) k (β−α)i
· f (k) (α) · iP
P =0
fn (α, β, t) = mn ( t ) · k! n (n+i)! (t−α)i
+
k =0 ·
i!·n! (β−α)i
i=0
n−k (157)
+i)! (β−t)i
n ( )k ( k
P (ni!·n! · i
−1 β −t) ( β−α )
· f (k) (β) · iP
P =0
qn ( t ) · k! n (n+i)! (β−t)i
k =0 · i!·n! (β−α)i
i=0
Adding and subtracting one from the bracketed terms, leads to the definitions for cn,k (t) and dn,k (t) as
specified by (42) and the required form for fn (α, β, t) as stated by (41).
n n
X (t − α) k X (t − β) k
· f (k ) ( α ) , · f (k ) ( β ) (159)
k! k!
k =0 k =0
β−α β−α
converge, respectively, over the intervals α < t < α + 2 and α + 2 < t < β. Consistent with (32),
a sufficient condition for convergence is for
(n)
fmax (β − α)n
lim = 0. (160)
n→∞ 2n · n!
It then follows that
n n
X (t − α) k X (t − β) k
· f (k) (α) · [1 + cn,k (t)] and · f (k) (β) · [1 + dn,k (t)] (161)
k! k!
k =0 k =0
n−k
(n+i)!
· δi
P
i!·n!
i=0 (t − α)
1 + cn,k (δ) = n
, δ= . (162)
(n+i)! (β − α)
· δi
P
i!·n!
i=0
It is clear that 1 + cn,k+1 (δ) contains one less term than 1 + cn,k (δ) in the numerator and all terms are
positive. Thus, 1 + cn,k+1 (δ) < 1 + cn,k (δ) and it is the case that 0 < 1 + cn,k (δ) ≤ 1 with equality for the
case of k = 0. For n and δ fixed, 1 + cn,k (δ) is a monotonically decreasing series for k ∈ {0, . . . , n}, as is
clearly evident in the graphs shown in Figure 5. It then follows, from Abel’s test for convergence [22],
that the series specified in (161) are convergent.
Second, the issue is the convergence of the spline approximation to the underlying function on
the interval [α, β]. This follows if lim cn,k (δ) = 0 for k ∈ {0, 1, . . . , n} as the summation is then that of
n→∞
a dual Taylor series. Further, mn (t) and qn (t) approach
ideal demarcation functions as n → ∞ . Thus,
the requirement is to prove converge of cn,k (δ) to zero for 0 ≤ δ ≤ 1/2. First, using the definition of
cn,k (t) specified by (42), it can readily be shown that
∂
h i
and it then follows that ∂δ −cn,k (δ) > 0 as the first two summations comprise only of positive terms
and the last double summation equals zero as the off-diagonal terms can be paired with one term being
the negative of the another. Thus, for n and k fixed
cn,k (δ1 ) < cn,k (δ2 ), (164)
when δ1 < δ2 . Hence, if lim cn,k (δ) = 0 for δ = 1/2, then the proof is complete. For δ = 1/2, it is the
n→∞
case that the denominator term for cn,k (1/2) simplifies according to
n
X (n + i)! 1
· = 2n (165)
i=0
i! · n! 2i
Math. Comput. Appl. 2019, 24, 35 34 of 40
assuming n k. Hence:
which clearly converges to zero as n increases for all fixed values of k. This completes the proof.
Rβ n f (k ) (α ) n−k
(n+i)! I1
1 P P
fn (α, β, λ)dλ = · · · +
α
(β−α)n+1 k =0
k!
i=0
i!·n! (β−α)i
(170)
n n−k
1 (−1)k (n+i)! I2
f (k ) ( β ) ·
P P
· · ·
(β−α)n+1 k =0
k!
i=0
i!·n! (β−α)i
where
Zβ Zβ
n+1 i+k
I1 = (β − α) · (λ − α) dλ, I2 = (λ − α)n+1 · (β − λ)i+k dλ. (171)
α α
λ−α
A change of variable r = β−α yields
Z1
n+2+i+k
I1 = ( β − α ) (1 − r)n+1 · ri+k dr = (β − α)n+2+i+k · B(i + k + 1, n + 2), (172)
0
Z1
n+2+i+k
I2 = ( β − α ) rn+1 · (1 − r)i+k dr = (β − α)n+2+i+k · B(n + 2, i + k + 1), (173)
0
As
Γ (m) Γ (n) (m − 1)! · (n − 1)!
B(m, n) = B(n, m) = = , m, n ∈ {1, 2, . . .}, (175)
Γ (m + n) (m + n − 1)!
it follows that
(n + 1)! · (i + k )!
I1 = I2 = (β − α)n+2+i+k · . (176)
(n + i + k + 2)!
Substitution yields the required result:
Rβ n
(β−α)k+1 n−k
(n+i)! (i+k )!
· [ f (k) (α) + (−1)k · f (k) (β)] ·
P P
fn (α, β, λ)dλ = (n + 1) · k! i! · (n+i+k +2)!
. (177)
α k =0 i=0
+1
2nP
(k ) (k )
fn (α, β, t) = di ( t − α ) i , fn (α, β, α) = f (k) (α), fn (α, β, β) = f (k) (β), k ∈ {0, 1, . . . , n} (178)
i=0
Zt n
X
fn (α, t, λ)dλ = cn,k · (t − α)k+1 · [ f (k) (α) + (−1)k f (k) (t)], (179)
α k =0
1
n=0 2
1 1
n=1 2 12
1 1 1
n=2 2 10 120
1 3 1 1
n=3 2 28 84 1680
1 1 1 1 1 . (180)
n=4 2 9 72 1008 30240
1 5 1 1 1 1
n=5 2 44 66 792 15840 665280
1 3 5 5 1 1 1
n=6 2 26 312 3432 11440 308880 17,297,280
k=0 k=1
(2n+1−k)!
The explicit expression cn,k = (n−k)n! ·
!(k+1)! 2 · (2n+1)!
can then be inferred. A direct proof of the equality
between this coefficient expression and the expression implicit in (177), i.e., a direct proof of
n−k
n + 1 X (n + i)! (k + i)! n! (2n + 1 − k)!
· = · (181)
k! i! (n + k + i + 2)! (n − k)!(k + 1)! 2 · (2n + 1)!
i=0
remains elusive.
f (t) f (α) n
cn,k (k + 1)(t − α)k · [ f (k) (α) + (−1)k f (k) (t)]+
P
f (t) = 2 + 2 +
k =1
n (182)
k +1 (1)
(−1)k f (k+1) (t) + Rn (α, t)
P
cn,k (t − α)
k =0
Math. Comput. Appl. 2019, 24, 35 36 of 40
− f (t) f (α) n
(1)
cn,k (k + 1)(t − α)k f (k) (α)+
P
−Rn (α, t) = 2 + 2 +
k =1 (183)
n
(t − α)k (−1)k f (k) (t)[(k + 1)cn,k − cn,k−1 ] + cn,n (−1)n (t − α)n+1 f (n+1) (t)
P
k =1
n+1
(k + 1)cn,k − cn,k−1 = −(k + 1)cn,k · (184)
n−k+1
Zt
eλ dλ = et − 1. (186)
0
Second, a spline based integral approximation, as specified by (49) and (51), for the integral of the
exponential function yields
Rt n
cn,k tk+1 1 + (−1)k et
h i
eλ dλ
P
≈
0 k =0 (187)
= [cn,0 t + cn,1 t2 + . . . + cn,n tn+1 ] + et [cn,0 t − cn,1 t2 + . . . + (−1)n cn,n tn+1 ]
Equating the right hand side of these two equations results in the required approximation
The result stated in (75), for the case of n → ∞ , arises from the result (54): c∞,k = 2k1+1 · (k+11)! .
The series convergence is based on the convergence of (49) for the case of an integrand of
f (t) = exp(t). A sufficient condition for this is convergence of a spline based approximation for exp(t)
tn (n)
over the interval [0, t]. Consistent with (47), a sufficient condition for this is for lim n ·f = 0.
n→∞ 2 · n! max
(n)
On the interval [0, t], fmax = et and, for t fixed:
tn (n) tn et
lim · f max = lim =0 (189)
n→∞ 2n · n! n→∞ 2n · n!
as required.
Math. Comput. Appl. 2019, 24, 35 37 of 40
Zt
ln(λ)dλ = t ln(t) − t + 1. (190)
1
The nth order based spline series approximation to the integral of ln(t), as defined by (49), is
Zt n
X
ln(λ)dλ ≈ cn,k (t − 1)k+1 [ f (k) (1) + (−1)k f (k) (t)], (191)
1 k =0
(−1)k+1 (k−1)!
where f (t) = ln(t) and f (k) (t) = tk
, k ∈ {1, 2, . . .}. Thus:
Zt n
(−1)k
X
k +1 k
ln(λ)dλ ≈ cn,0 (t − 1) ln(t) + (t − 1) cn,k (−1) (k − 1)!(t − 1) 1 + k . (192)
k =1
t
1
As cn,0 = 1/2, equating the right hand sides of this equation and (190) yields the approximation
n
(−1)k
(t − 1) X
k +1 k
t ln(t) − t + 1 ≈ · ln(t) + (t − 1) cn,k (−1) (k − 1)!(t − 1) 1 + . (193)
2
k =1
tk
n
k
2(t − 1) X ( −1 )
cn,k (−1)k+1 (k − 1)!(t − 1)k 1 +
ln(t) ≈ 1 + . (194)
t+1 tk
k =1
Zxo
1
dt = t1/h |xho = x1/h
o − z0 . (195)
ht1−1/h z0
zh0
With
k k
1 (−1)k Y 1 (−1)k+1 Y 1
f (t) = 1−1/h
, f (k ) ( t ) = k + 1−1/h
· ( j − ) = k + 1−1/h
· ( j − ), (196)
ht ht h t h
j=1 j=0
where the last expression for f (k) (t) is valid for k ∈ {1, 2, . . .}, it follows, based on a nth order spline
approximation as defined by (49) and (51), that
Rxo
" #
n k +1 k x1/h
1 k +1 1 1 k
(xo − zhi )
P Q o
dt ≈ cn,k (−1) (j − h ) · zkh+h−1 + (−1) ·
ht1−1/h xok+1
zhi k =0 j=0 i (197)
= ui+1 + νi+1 x1/h
o
Math. Comput. Appl. 2019, 24, 35 38 of 40
Here
n k +1 k
k +1
cn,k (−1) (xo − zhi ) j − h1 · kh+1h−1 ,
P Q
ui+1 =
k =0 j=0 zi
n
k
(198)
k +1
cn,k (xo − zhi ) j − h1 · k1+1
P Q
νi+1 = −
k =0 j=0 xo
As
Zxo
1
dt = xo1/h − zi ≈ ui+1 + νi+1 x1/h
o , (199)
ht1−1/h
zhi
zi + ui+1
xo1/h ≈ . (200)
1 − νi + 1
This approximation is zi+1 and is the basis for the next lower integral limit of zhi+1 . Thus:
zi + ui+1
747 Appendix 9. zi + 1 =
Proof: ArcCos, ArcSin and ArcTangent .
Approximation (201)
1 − νi+1
748 Consider a point ( x, y ) on the first quadrant of a unit circle, as illustrated in Figure 18, which defines an angle θ .
Appendix I. Proof: ArcCos, ArcSin and ArcTangent Approximation
d 1
749 The relationships θ ( y ) = asin ( y ) , and θ ( y ) = ------------------ , imply that the path length around the unit circle defined
d y
Consider a point (x, y) on the first quadrant 2 of a unit circle, as illustrated in Figure A1, which
1–y d
defines an angle
i
θ. The relationships θ ( y ) = asin ( y), and dy θ( y) = √ 1 2 , imply that the path length
750 by the angle of θ ⁄ 2 , to the point ( x o, y o ) , is 1−y
around the unit circle defined by the angle of θ/2i , to the point (xo , yo ), is
o y
Zyo
1
θ- = θ ------------------
751 ---
i i = 2p - dλ1 dλ, (202)
(202)
2 20 1 – λ 1 − λ2
0
752 where
where
θ π θ θ π θ
y θ = sin π = θ
cos − , x = θ
cos = π
sin θ − . (203)
753 y o = sin ---- = cos
o
2i --2- – ----i 2 2xio = cos ----i 2=i sin --2- –2----i 2i
o (203)
i
2 2 2 2
1 ( y o, x o )
2
θ⁄2
i y = 1–x
y ( x, y )
distance = θ
θ ( x o, y o )
i i
θ⁄2 distance = θ ⁄ 2
x 1
Figure 18. A1.
Figure Geometry, based
Geometry, on the
based on unit circle,
the unit for defining
circle, ( x o,(xyoo, ) ybased
for defining on on the angle θ/2i .
o ) based
i
the angle θ ⁄ 2 .
754 To relate y o to x and y , consider the well known half angle results for the first quadrant:
1 – cos ( θ ) sin ( θ )
tan ( θ ⁄ 2 ) = -------------------------- = --------------------------
1 + cos ( θ ) 1 + cos ( θ )
755 (204)
1 1
sin ( θ ⁄ 2 ) = ------- ⋅ 1 – cos ( θ ) cos ( θ ⁄ 2 ) = ------- ⋅ 1 + cos ( θ )
2 2
Math. Comput. Appl. 2019, 24, 35 39 of 40
To relate yo to x and y, consider the well known half angle results for the first quadrant:
q
1−cos(θ) sin(θ)
tan(θ/2) = 1+p
cos(θ)
= 1+cos(θ)
,
p (204)
sin(θ/2) = √1 · 1 − cos(θ), cos(θ/2) = √1 · 1 + cos(θ)
2 2
si−1 1 p 1 p
ti = si = √ · 1 − ci−1 , ci = √ · 1 + ci−1 ,
, (206)
1 + ci−1 2 2
where ti = tan θ/2i , si = sin θ/2i , ci = cos θ/2i , s0 = sin(θ) = y and c0 = cos(θ) = x. It then
follows that
yo
ti = tan(θ/2i ) = . (207)
xo
Using the notation gi (x) = ti , it follows that
v
t
q g2i (x)
2
yo = 1 − yo · gi (x) ⇒ yo = (208)
1 + g2i (x)
and
g (x)
q i
1 + g2 ( x )
Z i
i 1
θ=2 · p dλ (209)
0
1 − λ2
as required. The alternative definition for gi (x), as defined by (100), arises from a consideration of the
results for ti , i ∈ {0, 1, . . .}:
√
y
g0 ( x ) = x = 4x 1−x2
4x√2
= nd00 ,
y
g1 ( x ) = 2 1−x2
1+x = 2(1+x) q = nd1 ,
1
√ √
2− d0
g2 ( x ) = √2−2x
√ = √ = nd2 ,
2+ 2 1+x 2+ q d1 2
√ √ √ √ (210)
2− d
√2− 2√ 1√+x = √ = nd3 ,
1
g3 ( x ) =
2+q 2+ 2 1+x 2+ qd2 3
√ √ √ √
2− 2+ 2 1+x 2− d2
g4 ( x ) = q √ √ √ = √ = nd4
2 + d3 4
2+ 2+ 2+ 2 1+x
References
1. Burton, D.M. The History of Mathematics: An Introduction; McGraw-Hill: New York, NY, USA, 2005.
2. Katz, V.J. A History of Mathematics: An Introduction; Addison–Wesley: Boston, MA, USA, 2009.
3. Thomson, B.S. The natural integral on the real line. Scientiae Mathematicae Japonicae 2007, 67, 717–729.
4. Thomson, B.S. Theory of the Integral. Available online: http://classicalrealanalysis.info/com/documents/toti-
screen-Feb9-2013.pdf (accessed on 17 December 2018).
5. Botsko, M.W. A fundamental theorem of calculus that applies to all Riemann integrable functions. Math. Mag.
1991, 64, 347–348. [CrossRef]
6. Spivak, M. Calculus; Publish or Perish: Houston, TX, USA, 1994.
Math. Comput. Appl. 2019, 24, 35 40 of 40
7. Gradsteyn, I.S.; Ryzhik, I.M. Tables of Integrals, Series and Products; Academic Press: Cambridge, MA,
USA, 1980.
8. Horowitz, D. Tabular integration by parts. Coll. Math. J. 1990, 21, 307–311. [CrossRef]
9. Olver, F.W.J. Asymptotics and Special Functions; Academic Press: Cambridge, MA, USA, 1974; Chapter 1.
10. Stenger, F. The asymptotic approximation of certain integrals. SIAM J. Math. Anal. 1970, 1, 392–404.
[CrossRef]
11. Champeney, D.C. A Handbook of Fourier Theorems; Cambridge University Press: Cambridge, UK, 1987.
12. Pinkus, A. Weierstrass and Approximation Theory. J. Approx. Theory 2000, 107, 1–66. [CrossRef]
13. Davis, P.J.; Rabinowitz, P. Methods of Numerical Integration; Academic Press: Cambridge, MA, USA, 1984.
14. Howard, R.M. Antiderivative series for differentiable functions. Int. J. Math. Educ. Sci. Technol. 2004, 35,
725–731. [CrossRef]
15. Struik, D.J. A Source Book in Mathematics 1200–1800; Princeton University Press: Princeton, NJ, USA, 1986;
pp. 328–332.
16. Kountourogiannis, D.; Loya, P. A derivation of Taylor’s formula with integral remainder. Math. Mag. 2003,
76, 217–219. [CrossRef]
17. Baker, G.A. Essentials of Pade Approximants; Academic Press: Cambridge, MA, USA, 1975.
18. Finch, S.R. Mathematical Constants, Encyclopedia of Mathematics and its Applications; Cambridge University
Press: Cambridge, UK, 2003; Volume 94, pp. 56–57.
19. Weisstein, E.W. Pi Formulas, MathWorld—A Wolfram Web Resource. Available online: http://mathworld.
wolfram.com/PiFormulas.html (accessed on 17 December 2018).
20. Abramowitz, M.; Stegun, I.A. Handbook of Mathematical Functions with Formulas, Graphs and Mathematical
Tables; Dover: Mineola, NY, USA, 1965.
21. Spiegel, M.R.; Lipshutz, S.; Liu, J. Mathematical Handbook of Formulas and Tables; McGraw Hill: New York, NY,
USA, 2009.
22. Whittaker, E.T.; Watson, G.N. A Course of Modern Analysis; Cambridge University Press: Cambridge, UK, 1920.
© 2019 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).