See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341909807

PJM May2020 711 to 729

Article · June 2020

CITATIONS READS

0 139

2 authors:

Shyam Lal Indra Bhan

Banaras Hindu University Banaras Hindu University
96 PUBLICATIONS 310 CITATIONS 3 PUBLICATIONS 1 CITATION

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Indra Bhan on 04 June 2020.

The user has requested enhancement of the downloaded file.

Palestine Journal of Mathematics

Vol. 9(2)(2020) , 711–729 © Palestine Polytechnic University-PPU 2020

0
Approximation of functions with first and second derivatives f ,
00
f belonging to Lipschitz class of order 0 < α ≤ 1 by generalized
Legendre wavelet Expansion
Shyam Lal and Indra Bhan
Communicated by Ayman Badawi

MSC 2010 Classifications: 42C40, 65T60, 65L10, 65L60, 65R20

Keywords and phrases:Extended Legendre Wavelet, Extended Legendre Approximation, Function of Lipschitz class,
Orthonormal basis, Extended Legendre Wavelet Expansion.

Abstract In this paper , five new generalized Legendre wavelet estimators of functions having
first and second derivative belonging to Lipα [0, 1] have been established.

1 Introduction
Orthogonal functions have attracted the researchers of the modern analysis due to its
properties like optical control, applicability in a dynamical system. A function or signal of
L2 [0, 1) can be expressed in the form of wavelet expansion. There are several Fourier series
which is not convergent.Thus the difference of the function and nth partial sum of its expansion
can be calculated in some specific cases. Keeping this approach of Fourier analysis in mind, such
differences can be calculated in case of wavelet expansion in wavelet analysis. The whole
concept of approximation theory is based on Weierstrass approximation theorem. According
to this theorem, if a function f is continuous in [a,b] then there is a sequence (Bn (x))∞ n=0 of
Bernstein polynomials which converges uniformly to f on [a,b]. Legendre wavelet
approximates a function of Lipα class of order 0 < α ≤ 1 by piecewise Legendre polynomials.
Thus the Legendre wavelets defined on the interval [0,1), are generally obtained by a translation
operator on Legendre wavelet, defined on the subintervals[0, 21n ) of [0,1). In best of our
knowledge, till date nothing seems to have been done so for the degree of approximation of
0 00
the function f ∈ L2 [0, 1) whose first and second derivativesf andf belonging to Lipα [0, 1) of
order 0 < α ≤ 1 by Generalized Legendre wavelet methods. The purpose of this research paper
0 00
is to make an advanced study in this direction, five new wavelet estimators of f and f have
(µ)
been estimated by Ψn,m i.e. Generalized Legendre methods. Our estimators are better, sharper
and new in wavelet analysis and its applications.

2 Definitions and Preliminaries

Definition 2.1. Generalized or Extended LegendreSwavelet:
Let N denotes the set all natural numbers, N0 = N {0} and Zµ = {0, 1, 2, 3, ..., µ − 1} for a
positive integer µ. For a positive integer µ > 1, define the contracting mapping on the interval
I=[0,1] by
x+ξ
ψξ (x) = , x ∈ [0, 1], ξ ∈ Zµ .
µ
Then
[
ψξ (I ) ⊂ I, ∀ξ ∈ Zµ , ψξ (I ) = I.
ξ∈Zµ

The Legendre polynomials Pm (x) are defined by

1 dm
Pm (x) = [(x2 − 1)m ], m ≥ 0.
2m m! dxm
712 Shyam Lal and Indra Bhan

{Pm }∞
m=0 are orthogonal and satisfy :

P0 (t) = 1, P1 (t) = t

(m + 1)Pm+1 (t) = (2m + 1)tPm (t) − mPm−1 (t), m ∈ N.

Let
G0 = span {Pm (2t − 1)|t ∈ [0, 1], m ∈ Zµ } .
Let ξ ∈ Zµ , define Tξ on L2 [0, 1] by
( √
µf (ψξ−1 (t)) t ∈ ψξ (I )
Tξ f (t) =
0 , otherwise

Z 1
||Tξ (f )||22 = (Tξ f (x))2 dx
0
1
√
Z
= ( µf (ψξ−1 (x)))2 dx
0
Z 1
= µ f 2 (µx − ξ )dx
0
Z µ−ξ
= f 2 (v )dv, µx − ξ = v
−ξ

= ||f ||22
∴ ||Tξ (f )||22 = ||f ||22 .

Therefore, Tξ f is an isometry.
Suppose
M
Gk+1 = Tξ Gk , k ∈ N0 .
ξ∈Zµ

Then,

(i)Gk ⊂ Gk+1 , ∀ k ∈ N0 .
(ii)dimGk = M µk .
∞
[
(iii) Gk = L2 [0, 1].
k=0

Let n√ o
F0 = 2m + 1Pm (2t − 1)|t ∈ [0, 1], m ∈ ZM

. Then (i) F0 is an orthonormal

n o basis for G00 .
(ii)supp {Tξ f } supp Tξ f = φ, ξ 6= ξ ∀f ∈ L2 [0, 1],
T
0

where supp(f ) denotes the support of the function f.

For√m 6= m0 √
< 2m + 1Pm (2x − 1), 2m0 + 1Pm0 (2x − 1) >

√ √ Z 1
= 2m + 1 2m0 + 1 Pm (2x − 1)dxPm0 (2x − 1)dx
0
√ √ Z 1
= 2m + 1 2m0 + 1 Pm (y )dxPm0 (y )dy, 2x − 1 = y
−1
= 0, by property of Legendre polynomial.
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 713

√ √
< 2m + 1Pm (2x − 1), 2m + 1Pm (2x − 1) >
Z 1
2
= (2m + 1) Pm (2x − 1)dx
0
Z 1
2m + 1 2
= Pm (y )dy, 2x − 1 = y
2 −1
= 1.
n √ o
Fk = Tξ0 oTξ1 oTξ2 o...oTξk−1 ( 2m + 1Pm (2t − 1))|m ∈ ZM

√
F1 = {Tξ0 ( 2m + 1Pm (2t − 1)|m ∈ ZM )}
√ √
= { µ 2m + 1Pm (2ψξ−1 (x) − 1)|m ∈ ZM }
√ 1
= { 2m + 1µ 2 Pm (2µx − (2ξ + 1))|m ∈ ZM }
LetF1,m ∈ F1
Z 1
||F1,m ||22 = 2
(2m + 1)µPm (2µx − (2ξ + 1))dx
0
Z ξ+1
µ
2
= µ(2m + 1)Pm (2µx − (2ξ + 1))
ξ
µ
Z 1
2 du
= µ(2m + 1) Pm (u) , 2µx − (2ξ + 1) = u
−1 2µ
= 1.
< F1,m , F1,m0 > = 0, m 6= m0

Similarly ,
(
1, m = m0
< Fk,m , Fk,m0 > =
0, m 6= m0

Fk is an orthonormal basis for the vector space Gk , where “o” denotes composition of functions.[6]
Similarly, for n = 1, 2, 3, ..., µk , k ∈ N ,
( √ k 1 n
(µ) (µ) 2m + 1µ 2 Pm (2µk t − 2n + 1) t ∈ [ n− , )
µk µk
Ψn,m (t) = Ψ (k, m, n, t) =
0 , otherwise.
n o
(µ)
is called extended Legendre wavelet and Ψn,m is an orthonormal basis of Gk .
n o
(µ)
For µ = 2, Ψn,m reduces to known Legendre wavelet
 q
1 k n̂ − 1 n̂ + 1
m+ 2 2 Pm (2k t − n̂), ≤t<

Ψ(n,m
L)
(t) = 2 2k 2k
 0 , otherwise.
n o n o
(µ) (L)
Thus Ψn,m is a generalization of Ψn,m .

Definition 2.2. Extended Legendre approximation:

A function f ∈ L2 [0, 1] is expressed in the form of extended Legendre wavelet as
∞ X
X ∞ ∞ X
X ∞
f (t) = cn,m Ψ(n,m
µ)
(t) = < f, Ψ(n,m
µ)
> Ψ(n,m
µ)
(t) (2.1)
n=1 m=0 n=1 m=0
714 Shyam Lal and Indra Bhan

If the infinite series(2.1) is truncated, then it can be expressed as:

k
µ M
X X −1
f (t) = cn,m Ψ(n,m
µ)
(t) = C T Ψ(µ) (t) (2.2)
n=1 m=0

where C and Ψ(µ) (t) are µk M column vectors given by

C = [c1,0 , ..., c1,(M −1) , c2,0 , ..., c2,(M −1) , ..., cµk ,0 , ..., cµk ,(M −1) ]T ,

(µ) (µ) (µ) (µ) (µ) (µ)

and Ψ(µ) (t) = [Ψ1,0 (t), ..., Ψ1,(M −1) , Ψ2,0 (t), ..., Ψ2,(M −1) (t), ..., Ψµk ,0 (t), ..., Ψµk ,(M −1) (t)]T
respectively. We define
Z 1 !1/2
2
||f ||2 = |f (x)| dx
0
Pµk PM −1 (µ)
and Eµk ,M (f ) of f by (Sµk ,M f )(x) = n=1 m=0 cn,m Ψn,m (x)
is defined by

Eµk ,M (f ) = min kf − Sµk ,M (f )k2 , whereEµk ,M (f ) is the extended Legendre approximation.

Sµk ,M (f )

If Eµk ,M (f ) → 0 as n → ∞ then Eµk ,M (f ) is the best approximation of f .(Zygmund[1],pp.115).

Definition 2.3. Lipschitz class :
A function f ∈ Lipα [0, 1] if

|f (v + u) − f (v )| = O(|u|α ); 0 < α ≤ 1, v, u, u + v ∈ [0, 1][2].

i.e.
|f (v + u) − f (v )| = Mf |u|α , Mf ≥ 0, v, u, u + v ∈ [0, 1]

Example:

A functionf : [0, 1] → R is given by

f (x) = sinx
Consider,
   
2x + t t
|f (x + t) − f (x)| = |sin(x + t) − sinx| = 2cos sin ≤ |t|∀x, t, x + t ∈ [0, 1]
2 2
= O(|t|)
Then
f ∈ Lip1 [0, 1]

3 Theorems
We prove the following theorems:
0
Theorem 3.1. Let f ∈ L2 [0, 1) such that f ∈ Lipα [0, 1] and its extended Legendre wavelet
expansion be
X∞ X ∞
f (x) = cn,m Ψ(n,m
µ)
(x).
n=1 m=0

Then extended Legendre wavelet approximations satisfy:

∞ µk   
(µ) (1) (µ)
cn,0 Ψn,0 k2 = O µ1k 1 + 1
P P
(I)For f (x) = cn,0 Ψn,0 (x), Eµk ,0 (f ) = min kf − µkα
.
n=1 Sµk ,0 (f ) n=1
∞ ∞
P P (µ)
(II)For f (x) = cn,m Ψn,m (x),
n=1 m=0
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 715

k
µ M −1
(1) P P (µ)
Eµk ,M (f ) = min kf − cn,m Ψn,m k2
Sµk ,M (f ) n=1 m=0
  
1 1
=O 1 1 + µkα , M ≥ 2.
µk (2M −1) 2

00
Theorem 3.2. Let f ∈ L2 [0, 1) such that f ∈ Lipα [0, 1] and its expansion be

∞ X
X ∞
f (x) = cn,m Ψ(n,m
µ)
(x).
n=1 m=0

Then extended Legendre wavelet approximations satisfy:

∞ µk   
(µ) (2) (µ)
cn,0 Ψn,0 k2 = O µ1k 1 + 1
P P
(I)For f (x) = cn,0 Ψn,0 (x), Eµk ,0 (f ) = min kf − µk(α+1)
.
n=1 Sµk ,0 (f ) n=1
∞ 1
P P (µ)
(II)For f (x) = cn,m Ψn,m (x),
n=1 m=0
k
µ 1   
(2) P P (µ) 1 1
Eµk ,1 (f ) = min kf − cn,m Ψn,m k2 = O µ2k
1+ µkα
.
Sµk ,1 (f ) n=1 m=0
∞ ∞
P P (µ)
(III)For f (x) = cn,m Ψn,m (x),
n=1 m=0
k
µ M −1
(2) P P (µ)
Eµk ,M (f ) = min kf − cn,m Ψn,m k2
Sµk ,M (f ) n=1 m=0
  
1 1 1
=O 3 µ2k 1 + µkα , M ≥ 3.
(2M −3) 2

Proof 3.1. (I)

For

∞
(µ)
X
f (x) = cn,0 Ψn,0 (x)
n=1

(1) 0
The error en (x) between f (x) having f ∈ Lipα [0, 1] and its expression over any subinterval is defined as

 
(µ) n−1 n
e(1)
n (x) = cn,0 Ψn,0 (x) − f (x), x ∈ ,
µk µk
Z1
||e(1) 2
n ||2 = |e(1) 2
n (x)| dx
0
n
Zµk
= (f (x))2 dx − c2n,0 (3.1)
n−1
µk
716 Shyam Lal and Indra Bhan

Consider,
n 1
µk Zµk   2
n−1 n−1
Z
(f (x))2 dx = f +h dh, x = +h
µk µk
n−1 0
µk

1
Zµk     2
n−1 0 n−1
= f + hf + θh dh, 0 < θ < 1
µk µk
0
1
  2 Zµk   2
1 n−1 2 0 n−1
= f + h f + θh dh
µk µk µk
0
1
 Z µk  
n−1 0 n−1
+ 2f hf + θh dh.
µk µk
0
Next,
(µ)
cn,0 = < f, Ψn,0 >
n
Zµk
k
= f (x)µ 2 dx
n−1
µk

1
Zµk  
k n−1 n−1
= µ2 f + h dh, x = +h
µk µk
0
1
Zµk     
k n−1 0 n−1
cn,0 = µ2 f + hf + θh dh, 0 < θ < 1
µk µk
0
 1

  Zµk  
k  1 n−1 0 n−1 
= µ2 
 µk f + hf + θh dh 
µk µk 
0

Now,
 1 2
  2 Zµk    
1 n−1 0 n−1
c2n,0 + µk 
 
= k
f k
h f k
+ θh dh
µ µ  µ 
0

1
  Zµk  
n−1 0 n−1
+ 2f hf + θh dh.
µk µk
0
Then,
1
 2 2
Zµk   2 Zµk  
0 n−1 0 n−1
||e(1) 2
h2 f dh − µk 
 
n ||2 = k
+ θh hf k
+ θh dh
µ  µ 
0 0

= I1 − I2 , say. (3.2)
Consider,
1
Zµk   2
2 0 n−1
I1 = h f + θh dh
µk
0
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 717

1
Zµk       2
0 n−1 0 n−1 0 n−1
I1 = h2 f + θh − f + f dh
µk µk µk
0

1 1
Zµk     2 Zµk   2
0 n−1 0 n−1 0 n−1
= h2 f + θh − f dh + h f2
µk µk µk
0 0
1
µk        
n−1 n−1 n−1
Z
0 0 0
+ 2 h2 f + θh − f f dh
µk µk µk
0
1 1 1
µk
Z Zµk Zµk
|I1 | ≤ M12 h2α+2 dh + M22 h2 dh + 2M1 M2 hα+2 dh, f ∈ Lipα [0, 1]
0 0 0

M12 M22 2M1 M2

= + + (3.3)
(2α + 3)µk(2α+3) 3.µ3k (α + 3)µk(α+3)
Also,
 1
2
Zµk  
0 n−1
= µk 
 
I2 hf k
+ θh dh
 µ 
0

1
 2
Zµk       
0 n−1 0 n−1 0 n−1
= µk 
 
h f + θh − f + f dh dh
 µk µk µk 
0

 1 1
2
Zµk     Zµk  
0 n−1 0 n−1 0 n−1
µk 
 
|I2 | ≤ h f + θh − f dh + h f dh
 µk µk µk 
0 0

 1 2
2
Zµk Zµk
0 0
µk  hM1 hα dh + M2
 
≤  hdh

0 0

0 0
!2
M1 M2
= µk +
(α + 2)µk(α+2) 2µ2k
0 0 0 0
M1 2 M2 2 M1 M2
= (2 +3)
+ 3k
+ (3.4)
2
(α + 2) µ k α 4 µ (α + 2)µk(α+3)
By eqns (3.2), (3.3) and (3.4) , we have
M12 M22 2M1 M2
||e(1) 2
n ||2 ≤ + +
(2α + 3)µk(2α+3) 3µ3k (α + 3)µk(α+3)
0 0 0 0
M1 2 M2 2 M 1 M2
+ + +
(α + 2)2 µk(2α+3) 4µ3k (α + 2)µk(α+3)
N12 7N22
   
1 1 2N1 N2 1 1
≤ + + + +
µk(2α+3) (2α + 3) (α + 2)2 12µ3k µk(α+3) (α + 2) 2(α + 3)
 2
1 N1 N2 0 0
≤ + , N1 = max[M1 , M1 ], N2 = max[M2 , M2 ]
µk µk(α+1) µk
718 Shyam Lal and Indra Bhan

2
Z1
 k
µ
(1)
X
Now, (Eµk ,0 (f ))2 =  e(1)
n (x)
 dx
0 n=1

Z1 X
µk µk Z1
µk X
(1)
X
= (e(1) 2
n (x)) dx +2 e(1)
n (x)en0 (x)dx
0 n=1 n=1 n6=n0 0

µ Z k 1
X
= (en (x))2 dx, due to disjoint supports of en and e0n
n=1 0

k
µ
X
= ||e(1) 2
n ||2
n=1

µk  2
X 1 N1 N2
≤ + k
µk µk(α+1) µ
n=1
 2
N1 N2
= + k .
µk(α+1) µ
 2
2 1 1
≤ N + k(α+1) , whereN = max[N1 , N2 ]
µk µ
Therefore,
 
(1) 1 1
Eµk ,0 (f ) ≤ N + k(α+1)
µk µ
  
1 1
= O 1 + .
µk µkα
(II)
For
∞ X
X ∞
f (x) = cn,m Ψ(n,m
µ)
(x).
n=1 m=0
n
Zµk √ k
cn,m = f (x) 2m + 1µ 2 Pm (2µk x − n̂)dx
n−1
µk

√ Z1  
2m + 1 n̂ + t
= k f Pm (t)dt, 2µk x − n̂ = t
2µ 2 2µk
−1

√ Z1    
2m + 1 n̂ + t Pm+1 (t) − Pm−1 (t)
= k f d
2µ 2 2µk 2m + 1
−1
   1
1 n̂ + t
= k 1 f (P m+1 (t) − P m−1 (t))
2µ 2 (2m + 1) 2 2µk −1

Z1  
1 0 n̂ + t
− 3k 1 f (Pm+1 (t) − Pm−1 (t))dt, integrating by parts.
4µ 2 (2m + 1) 2 2µk
−1

Z1       
−1 0 n̂ + t 0 n̂ 0 n̂
= 3k 1 f −f +f (Pm+1 (t) − Pm−1 (t)
4µ 2 (2m + 1) 2 2µk 2µk 2µk
−1
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 719

Z1     
−1 0 n̂ + t 0 n̂
cn,m = 3k 1 f −f (Pm+1 (t) − Pm−1 (t))dt
4µ 2 (2m + 1) 2 2µk 2µk
−1

Z1  
1 0 n̂
− 3k 1 f (Pm+1 (t) − Pm−1 (t))dt
4µ 2 (2m + 1) 2 2µk
−1

Z1    
1 0 n̂ + t 0 n̂
|cn,m | ≤ 3k 1 f −f |Pm+1 (t) − Pm−1 (t)|dt
4µ 2 (2m + 1) 2 2µk 2µk
−1

Z1  
1 0 n̂
+ 3k 1 f |Pm+1 (t) − Pm−1 (t)|dt
4µ 2 (2m + 1) 2 2µk
−1

Z1   
1 Q1
≤ 3k 1 + Q2 |Pm+1 (t) − Pm−1 (t)|dt
4µ (2m + 1) 2
2 µkα
−1

   Z1
1 Q1
= 3k 1 + Q2 |Pm+1 (t) − Pm−1 (t)|dt
4µ (2m + 1) 2
2 µkα
−1

Since,
 21  1  21
Z1 Z1

Z
|Pm+1 (t) − Pm−1 (t)|dt ≤  12 dt  |Pm+1 (t) − Pm−1 (t)|2 dt
−1 −1 −1
 1  21
√ Z
2 2
= 2  (Pm +1 (t) − Pm−1 (t))dt

−1
 12
√

2 2
= 2 +
2m − 1 2m + 3
√
2 2
≤ √ ,
2m − 1
therefore
   √
1 Q1 2 2
|cn,m | ≤ 3k 1 + Q2 √
4µ (2m + 1) 2
2 µkα 2m − 1
 
1 Q1
≤ √ 3k + Q2 . (3.5)
2µ (2m − 1)
2 µkα
 k k

µ −1 ∞ µ M −1
M
!
X X X X X
f (x) − Sµk ,M (f ) =  + cn,m Ψ(n,m
µ)
(x) − cn,m Ψ(n,m
µ)
(x)
n=1 m=0 m=M n=1 m=0
k
µ
X ∞
X
= cn,m Ψ(n,m
µ)
(x)
n=1 m=M

(Eµ1 k ,M (f ))2 = min kf − Sµk ,M (f )k22

Sµk ,M (f )
Z 1
= |f (x) − Sµk ,M (f )|2 dx
0
 k 2
Z 1 Xµ ∞
X
=  cn,m Ψ(n,m
µ)
(x) dx
0 n=1 m=M
720 Shyam Lal and Indra Bhan

 k
Z 1 Xµ ∞
X
(µ)2
(Eµ1 k ,M (f ))2 =  c2n,m Ψn,m (x)
0 n=1 m=M

(µ)
XXXX
+ 2 cn,m cn0 m0 Ψ(n,m
µ)
(x)Ψn0 m0 (x) dx
1≤n6=n0 ≤µk ,M ≤m6=m0 <∞
k
µ
X ∞
X Z 1
(µ)2
= c2n,m Ψn,m (x)dx
n=1 m=M 0
Z 1
(µ)
XXXX
+ 2 cn,m cn0 m0 Ψ(n,m
µ)
(x)Ψn0 m0 (x)dx
1≤n6=n0 ≤µk ,M ≤m6=m0 <∞ 0

k
µ
X ∞
X
= c2n,m kΨ(n,m
µ) 2 (µ)
k2 , other term vanish due to orthonormality of ψn,m .
n=1 m=M
k
µ
X ∞
X
= |cn,m |2
n=1 m=M

Then,
k
µ ∞
(1)
X X
(Eµk ,M (f ))2 = min kf − Sµk ,M (f )k22 = |cn,m |2
Sµk ,M (f )
n=1 m=M

µ ∞ k  !2
X X 1 Q1
≤ √ 3k 1 + Q2 , by eqn(3.5)
2µ 2 (2m − 1) 2 µkα
n=1 m=M
µk
2  2 ∞
X Q1 1 X 1
= + Q2
µkα 2µ3k (2m − 1) m=M
n=1
2 Z∞
µk

Q1 1 1
= + Q2 dm
2 µkα 2µ3k (2m − 1)2
M
 2
Q1 1 1
= kα
+ Q2 2k
µ 4µ (2M − 1)
2
 2
Q 1 1
≤ 2 k
1 + kα , Q = max[Q1 , Q2 ].
4µ µ (2M − 1)
Thus,
 
(1) Q 1 1
Eµk ,M (f ) ≤ 1+ 1 .
2µk µkα (2M − 1) 2
Hence,
 !
(1) 1 1 1
Eµk ,M (f ) = O 1 1 + kα .
(2M − 1) 2 µk µ

Thus, the Theorem (3.1)is completely established.

Proof 3.2. (I)

(2) 00
The error en (x) between f (x)having f ∈ Lipα [0, 1] and its expression over any subinterval is
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 721

defined as
 
(µ) n−1 n
e(2)
n (x) = cn,0 Ψn,0 (x) − f (x), x ∈ ,
µk µk
n
Zµk
||e(2) 2
n ||2 = (f (x))2 dx − c2n,0 . (3.6)
n−1
µk

Consider,
n 1
µk Zµk   2
n−1 n−1
Z
(f (x))2 dx = f +h dh, x = +h
µk µk
n−1 0
µk

1
Zµk   2
h2 00 n − 1
   
n−1 0 n−1
= f + hf + f + θh dh, 0 < θ < 1
µk µk 2 µk
0
1
  2   2 Zµk   2
1 n−1 1 0 n−1 1 4 00 n−1
= f + 3k f + h f + θh dh
µk µk 3µ µk 4 µk
0
1
     Z µk  
1 n−1 0 n−1 n−1 2 00 n−1
+ f f +f h f + θh dh
µ2k µk µk µk µk
0
1
 Z µk  
0 n−1 00 n−1
+ f h3 f + θh dh (3.7)
µk µk
0
(µ)
cn,0 = < f, Ψn,0 >
1
Zµk  
k n−1 n−1
= µ 2 f + h dh, x = +h
µk µk
0
1
Zµk  
h2 00 n − 1
    
k n−1 0 n−1
= µ 2 f + hf + f + θh dh, 0 < θ < 1
µk µk 2 µk
0
 1

    µk  
n−1 n−1 n−1
Z
k  1 1 0 1 00
h2 f

= µ2 
 µk f + f + + θh dh
.
µk 2µ2k µk 2 µk
0

Now,
 2 2
2 2 µk
µk 
       
n−1 n−1  h2 f 00 n − 1 + θh dh 
Z
1 1 0
c2n,0

= f + f +
µk µk 4µ3k µk 4  µk 
0

1
      Zµk  
1 n−1 0 n−1 n−1 00 n−1
+ f f +f h2 f + θh dh
µ2k µk µk µk µk
0
1
 Z µk  
1 0 n−1 00 n−1
+ f h2 f + θh dh (3.8)
2µk µk µk
0
722 Shyam Lal and Indra Bhan

By eqns (3.6), (3.7)and (3.8), we have

1
  2 Zµk   2
1 0 n−1 1 00 n−1
||e(2) 2
n ||2 = f + 4
h f + θh dh
12µ3k µk 4 µk
0
1
 1 2
 Zµk µk
µk 
      
n−1 n−1  h2 f 00 n − 1 + θh dh 
Z
0 00
h3 f

+ f + θh dh −
µk µk 4  µ k 
0 0

2
  Zµk  
1 0 n−1 2 00 n−1
− f h f + θh dh
2µk µk µk
0
= I1 + I2 + I3 − I4 − I5 , say (3.9)
  2
1 0 n−1
I1 = f
12µ3k µk
R12
|I1 | ≤ . (3.10)
12µ3k
1
Zµk   2
1 4 00 n−1
I2 = h f + θh dh
4 µk
0
1
Zµk       2
1 4 00 n−1 00 n−1 00 n−1
= h f + θh − f +f dh
4 µk µk µk
0
1 2
Zµk    2 Zµk  2
1 00 n−1 00 n−1 1 00 n−1
|I2 | ≤ h4 f + θh − f dh + 4
h f dh
4 µk µk 4 µk
0 0
1
µk      
n−1 n−1 n−1
Z
1 00 00 00
+ h4 f + θh − f f dh
2 µk µk µk
0

R22 R32 R2 R3 2α
≤ + + . (3.11)
4(2α + 5)µk(2α+5) 20µ5k 2(α + 5)µk(α+5)
1
  Zµk  
0 n−1 00 n−1
I3 = f h3 f + θh dh
µk µk
0
R1 R2 R1 R3
|I3 | ≤ + (3.12)
(α + 4)µ k (α +4) 4µ4k
 1 2
µk
µk 
   
 h2 f 00 n − 1 + θh dh 
Z

I4 = k
4  µ 
0

R22 R32 2R2 R3

|I4 | ≤ (2 +5)
+ 5k
+ . (3.13)
2
4(α + 3) 2 k α 36µ 12(α + 3)µk(α+5)
2
  Zµk  
1 0 n−1 2 00 n−1
I5 = f h f + θh dh
2µk µk µk
0
R1 R2 R1 R3
|I5 | ≤ + . (3.14)
2(α + 3)µk(α+4) 6µ4k
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 723

By eqns (3.9) to (3.14), we have

R12 R22 22α R32 R2 R3 R1 R2

||e(2) 2
n ||2 ≤ + + + +
12µ3k (2α + 5)µk(2α+5) 20µ5k 2(α + 5)µk(α+5) (α + 4)µk(α+4)
R1 R3 R22 R32 2R2 R3 R1 R2
+ + + + +
4µ4k 4(α + 3)2 µk(2α+5) 36µ5k 12(α + 3)µk(α+5) 2(α + 3)µk(α+4)
R1 R3
+
6µ4k
R12 R22 R32
   
1 1 1 1
= + + + +
12µ3k 4µk(2α+5) 2α + 5 (α + 3)2 4µ5k 5 6
   
R1 R2 1 1 R1 R3 1 1
+ + + +
µk(α+4) α + 4 2(α + 3) 2µ4k 2 3
 
R2 R3 1 1
+ +
µk(α+5) (α + 5) (α + 3)
R22 R32
 
1 2 2R1 R2 2R1 R3 2R2 R3
≤ R 1 + + + + +
µ3k µ2k(α+1) µ2k µk(α+1) µk µk(α+2)
 2
1 R2 R3
||e(2) 2
n ||2 ≤ R1 + + . (3.15)
µ3k µk(α+1) µk
Next,
2
Z1
 k
µ
(2)
X
(Eµk ,0 (f ))2 =  e(2)
n (x)
 dx
0 n=1

Z1 X
µk µk Z1
µk X
(2)
X
(2)
= 2
(en (x)) dx + 2 e(2)
n (x)en0 (x)dx
0 n=1 n=1 n6=n0 0

µ Z k 1
X
0
= (e(2) 2
n (x)) dx, due to disjoint supports of en and en
n=1 0

k
µ
X
= ||e(2) 2
n ||2
n=1

µk  2
X 1 R2 R3
≤ R1 + k(α+1) + k , by eqn (3.15)
µ3k µ µ
n=1
 2
k 1 R2 R3
= µ 3k R1 + k(α+1) + k .
µ µ µ
Then,
 
(2) 2R 1
Eµk ,0 (f ) ≤ 1 + k(α+1) , R = max[R1 , R2 , R3 ]
µk µ
  
1 1
= O 1 + .
µk µk(α+1)
Thus,
  
(2) 1 1
Eµk ,0 (f ) = O 1+ .
µk µk(α+1)
724 Shyam Lal and Indra Bhan

(II)
k
µ X
X 1
f (x) = cn,m Ψ(n,m
µ)
(x).
n=1 m=0

(3) 00
The error en (x) between f (x) having f ∈ Lipα [0, 1] and its expression over any subinterval
is defined as

(µ) (µ)
e(3)
n (x) = cn,0 Ψn,0 (x) + cn,1 Ψn,1 − f (x)
n
Zµk
||e(3) 2
n ||2 = (f (x))2 dx − c2n,0 − c2n,1 (3.16)
n−1
µk

Now, consider
n

√ Zµk
k
cn,1 = 3µ 2 f (x)P1 (2µk x − n̂)dx
n−1
µk

√ Zµk  
k n−1
= 3µ 2 f + h (2µk h − 1)dh
µk
0
1

√ Zµk  
h2 00 n − 1
    
k n−1 0 n−1
= 3µ 2 f + hf + f + θh (2µk h − 1)dh
µk µk 2 µk
0

 1

√

   Zµk  

k
 1 0 n−1 1 k 2 00 n−1 
cn,1 = 3µ 2 f + (2µ h − 1)h f + θh . (3.17)

 6µ2k µk 2 µk 

 0 

By eqns (3.7), (3.8),(3.16)and (3.17),we have

1
 1 2
Zµk  2 µk
µk 
   
n−1  h2 f 00 n̂ − 1 + θh dh
Z
1 00
||e(3) 2
h2 f

n ||2 = + θh dh −
4 µk 4  µk 
0 0

1
 2
Zµk  
3 k 00 n−1
µ  (2µk h − 1)h2 f

− + θh dh
4  µk 
0

= I1 − I2 − I3 , say (3.18)
1
µk    2
n̂ − 1
Z
1 2 00
I1 = h f + θh dh
4 µk
0

T12 T22 T1 T2
|I1 | ≤ + + . (3.19)
4(2α + 5)µ k (2 α+5) 20µ5k 2(α + 5)µk(α+5)
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 725

 1
2
Zµk
µk 
 
00 n−1
h2 f

I2 =  + θh dh
4  µk 
0

T12 T22 T1 T2
|I2 | ≤ + + . (3.20)
4(α + 3)2 µk(2α+5) 36µ5k 2(α + 5)µk(α+5)
 1 2
Zµk  
3 k 00 n−1
µ  (2µk h − 1)h2 f

I3 = + θh dh
4  µk 
0

27T12 75T22 15T1 T2

|I3 | ≤ + + . (3.21)
2
4(α + 3) .µ k (2α+5) 144µ5k 4(α + 3)µk(α+5)
By eqns(3.18) to (3.21), we have

 2
29 T1
||e(3) 2
n ||2 ≤ + T2
µ5k µkα
2
29T 2

1
= 1 + kα , T = max[T1 , T2 ] (3.22)
µ5k µ

Then,
k
µ
(2)
X
(Eµk ,1 (f ))2 = ||e(3) 2
n ||2
n=1
2
29T 2

1
≤ 1 + kα , by (3.22)
µ4k µ
√  
(2) 29T 1
Eµk ,1 (f ) ≤ 1 + kα
µ2k µ
  
1 1
= O 1 + .
µ2k µkα
Hence,
  
(2) 1 1
Eµk ,1 (f ) = O 1 + .
µ2k µkα
(III)
For ∞ X
∞
X
f (x) = cn,m Ψ(n,m
µ)
(x).
n=0 m=0

Following the proof of the second part of the theorem (3.1),

Z1  
1 1 0 n̂ + t
cn,m = √ 3k f Pm−1 (t) − Pm+1 (t)dt
4 2m + 1 µ 2 2µk
−1

Z1   
1 1 0 n̂ + t d(Pm (t) − Pm−1 (t))
= √ 3k f
4 2m + 1 µ 2 2µk 2m − 1
−1

Z1  
1 1 0 n̂ + t d(Pm+2 (t) − Pm (t))
− 3k f
2µk
p
(2m + 1) µ 2 2m + 3
−1
726 Shyam Lal and Indra Bhan

Z1   
1 1 00 n̂ + t (Pm (t) − Pm−1 (t))
cn,m = √ 5k f
8 2m + 1 µ 2 2µk 2m − 1
−1

Z1  
1 1 00 n̂ + t (Pm+2 (t) − Pm (t))
− √ 5k f
8 2m + 1 µ 2 2µk 2m + 3
−1

Z1  
1 1 00 n̂ + t τ (t)
= √ 5k f dt
8 2m + 1 µ 2 2µk (2m − 1)(2m + 3)
−1

where τ (t) = (2m + 3)Pm−2 (t) − 2(2m + 1)Pm (t) + (2m − 1)Pm+2 (t).
Then,
Z1  
1 1 00 n̂ + t
|cn,m | = √ 5k f τ (t) dt
8(2m − 1)(2m + 3) 2m + 1 µ 2 2µk
−1

Z1  
1 1 00 n̂ + t
≤ √ 5k f |τ (t)| dt
8(2m − 1)(2m + 3) 2m + 1 µ 2 2µk
−1

Consider,
Z1   Z1      
00 n̂ + t 00 n̂ + t 00 n̂ 00 n̂
f |τ (t )| dt = f − f + f |τ (t)| dt
2µk 2µk 2µk 2µk
−1 −1

Z1     Z1  
00 n̂ + t 00 n̂ 00 n̂
≤ f −f |τ (t)| dt + f |τ (t)| dt
2µk 2µk 2µk
−1 −1

Z1  
A1
≤ + A 2 |τ (t)|dt
µkα
−1
 21  1  12
Z1

  Z
A1
≤ + A2  12 dt  |τ (t)|2 dt
µkα
−1 −1
√
 
A1 2m + 3
≤ 24 + A2 √
µkα 2m − 3
q
3  
8 A1
|cn,m | ≤ 5k + A 2 . (3.23)
µ 2 (2m − 3)2 µkα
Then,
k
µ ∞
(2)
X X
(Eµk ,M (f ))2 = |cn,m |2
n=1 m=M
k
µ ∞  2
X X 3 1 A1
≤ + A2 , by (3.23)
8µ (2m − 3)4
5k µkα
n=1 m=M
k
µ  2 ∞
X A1 3 X 1
= + A 2 5
µkα 8µ k (2m − 3)4 m=M
n=1
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 727

 2 Z∞
(2) 1 A1 dm
(Eµk ,M (f ))2 = + A2 .
16µ4k µkα (2m − 3)4
M

Thus,
 
(2) 1 A1 1
Eµk ,M (f ) ≤ + A 2 3
4µ2k µ kα
(2m − 3) 2
 
A 1
≤ 3 1 + , A = max[A1 , A2 ].
4µ2k (2M − 3) 2 µkα

Hence,
 !
(2) 1 1 1
Eµk ,M (f ) = O 3 1 + kα , M ≥ 3.
(2M − 3) 2 µ2k µ

Thus, the Theorem (3.2) is completely proved.

4 Corollaries
Following corollaries are deduced from Theorem (3.1) and (3.2)

0
Corollary 4.1. Let f ∈ L2 [0, 1) such that f ∈ Lipα [0, 1] and its Legendre wavelet expansion be

X ∞
∞ X ∞ X
X ∞
f (x) = cn,m Ψ(n,m
L)
(x) = < f, Ψ(n,m
L)
> Ψ(n,m
L)
(x).
n=1 m=0 n=1 m=0

Then Legendre wavelet approximations satisfy:

∞
P (L)
(i)For f (x) = cn,0 Ψn,0 (x),
n=1
k−1
2P
(1) (L) 1 1



E2k−1 ,0 (f ) = min kf − cn,0 Ψn,0 k2 = O 2k 1+ 2kα .
S2k−1 ,0 (f ) n=1
∞ P ∞
P (L)
(ii)For f (x) = cn,m Ψn,m (x),
n=1 m=0
k−1
2P M −1
(1) P (L)
E2k−1 ,M (f ) = min kf − cn,m Ψn,m k2
S2k−1 ,M (f ) n=1 m=0
 
1 1


=O 1 1 + 2kα , M ≥ 2.
2k (2M −1) 2

00
Corollary 4.2. Let f ∈ L2 [0, 1) such that f ∈ Lipα [0, 1] and its extended Legendre wavelet
expansion be
∞ X
X ∞ ∞ X
X ∞
f (x) = cn,m Ψ(n,m
L)
(x) = < f, Ψ(n,m
L)
> Ψ(n,m
L)
(x).
n=1 m=0 n=1 m=0

Then extended Legendre wavelet approximations satisfy:

∞
P (L)
(i)For f (x) = cn,0 Ψn,0 (x),
n=1
k−1
2P
(2) (L) 1 1



E2k−1 ,0 (f ) = min kf − cn,0 Ψn,0 k2 = O 2k 1+ 2k(α+1)
.
S2k−1 ,0 (f ) n=1
∞ P
1
P (L)
(ii)For f (x) = cn,m Ψn,m (x),
n=1 m=0
k−1
2P 1
(2) P (L) 1 1



E2k−1 ,1 (f ) = min kf − cn,m Ψn,m k2 = O 22k
1+ 2kα .
S2k−1 ,1 (f ) n=1 m=0
728 Shyam Lal and Indra Bhan

∞ P
∞
P (L)
(iii)For f (x) = cn,m Ψn,m (x),
n=1 m=0
k−1
2P M −1
(2) P (L)
E2k−1 ,M (f ) = min kf − cn,m Ψn,m k2
S2k−1 ,M (f ) n=1 m=0
 
1 1 1


=O 3 22k 1 + 2kα , M ≥ 3.
(2M −3) 2

Proofs of Corollaries (5.1) and (5.2)are followed by the proof of Theorem (3.1) and (3.2)
respectively.

5 Conclusions
(1)The estimates of the Theorems  are obtained as following
(1) 1 1 (1)
(i)Eµk ,0 (f ) = O µk 1 + µkα .Eµk ,0 (f ) → 0 as k → ∞
  
(1) 1 1 (1)
(ii)Eµk ,M (f ) = O 1 1 + µkα
.Eµk ,M (f ) → 0 as k → ∞, M → ∞
(2M −1) 2 µk
  
(2) (2)
(iii)Eµk ,0 (f ) = O µ1k 1 + µk(1α+1) .Eµk ,0 (f ) → 0 as k → ∞
  
(2) (2)
(iv)Eµk ,1 (f ) = O µ12k 1 + µkα 1
.Eµk ,1 (f ) → 0 as k → ∞
  
(2) 1 1 1
(v)Eµk ,M (f ) = O 3 µ2k 1 + µkα .
(2M −3) 2
(2)
Eµk ,M (f ) → 0 as k → ∞, M → ∞.
(1) (1) (2) (2) (2)
Therefore Eµk ,0 (f ), Eµk ,M (f ), Eµk ,0 (f ), Eµk ,1 (f )andEµk ,M (f ) are best possible errors of ap-
proximations in wavelet analysis.
00
(2) Generalised Legendre wavelet estimators of f belonging to Lipα [0, 1] is better and sharper
0
than the estimator of f belonging to Lipα [0, 1].
(1) (1)
(3)Eµk ,0 (f ) can not be obtained directly by Eµk ,M (f ) by taking M = 0.
(2) (2) (2)
(4)Eµk ,0 (f ) and Eµk ,1 (f ) also not obtained by Eµk ,M (f ) by taking M = 0 and M = 1 respec-
(1) (1) (2) (2) (2)
tively. Hence,Eµk ,0 (f ), Eµk ,M (f ), Eµk ,0 (f ), Eµk ,1 (f ) and Eµk ,M (f )are estimated separately.
(1) (1) (2) (2) (2)
(5) Legendre wavelet errors of approximations E2k−1 ,0 (f ), E2k−1 ,M (f ), E2k−1 ,0 (f ), E2k−1 ,1 (f )andE2k−1 ,M (f )
(1) (1) (2) (2) (2)
are obtained by Eµk ,0 (f ), Eµk ,M (f ), Eµk ,0 (f ), Eµk ,1 (f )and Eµk ,M (f ) respectively by taking
µ = 2.

References
[1] A. Zygmund , Trigonometric Series Volume I, Cambridge University Press, 1959.
[2] E. C. Titchmarsh, The Theory of functions, Second Edition, Oxford
University Press, (1939).
[3] R.A.Devore, Nonlinear approximation, Acta Numerica, Vol.7, Cambridge University Press, Cambridge,
1998, pp. 51-150.
[4] J. Morlet, G. Arens, E. Fourgeau and D. Giard, Wave propagation and
sampling theory, part I; Complex signal and scattering in multilayer media, Geophysics 47(1982) No. 2,
203-221.
[5] J. Morlet, G. Arens, E. Fourgeau and D. Giard, Wave propagation and
sampling theory, part II; sampling theory complex waves, Geophysics 47(1982) no. 2, 222-236.
[6] C.A.Micchelli and Y. Xu, Using the matrix refinement equation for the construction of wavelets on invari-
ant set, Appl. Comput. Harmon. Anal.,1(1994), 391-401.
[7] R. A. Devore, Nonlinear Approximation, Acta Numerica, Vol. 7, Cambridge University
Press,Cambridge(1998), pp. 51-150.
[8] Shyam Lal and Susheel Kumar “Best Wavelet Approximation of function belonging to Generalized Lips-
chitz Class using Haar Scaling function,” Thai Journal of Mathematics, Vol. 15(2017), No. 2, pp. 409-419.
[9] Xu, Xiaoyong, and Da Xu. “Legendre wavelets method for approximate solution of fractional-order differ-
ential equations under multi-point boundary conditions." International Journal of Computer Mathematics
(2017): 1-17.
 0 00 (µ)
Approximation of functions with f , f ∈ Lipα [0, 1]by Ψn,m 729

Author information
Shyam Lal and Indra Bhan, Department of Mathematics, Institute of Science,
Banaras Hindu University, Varanasi-221005,, India.
E-mail: Shyam Lal <[email protected]> and Indra Bhan <[email protected]>

Received: May 7, 2018.

Accepted: January 27, 2019.

View publication stats