COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)

ISSN:2320-0790

Error Estimations in an Approximation on a Compact

Interval with a Wavelet Bases
Dr. Marco Schuchmann1 , Michael Rasguljajew2
1,2

Hochschule Darmstadt
Abstract: By an approximat ion with a wavelet base we have in practise not only an error if the function y is not in Vj .
There we have a second error because we do not use all bases functions. If the wavelet has a compact support we have no
error by using only a part of all basis function. If we need an approximation on a compact interval I (which we can do even
if y is not quadratic integrable on R, because in that case it must only be quadratic integrab le on I) leads to worse
approximations if we calculate an orthogonal projection from 1 I y in Vj . We can get much better approximat ions, if we apply
a least square approximat ion with points in I. Here we will see, that this approximation can be much better th an a orthogonal
projection form y or 1 I y in Vj . With the Shannon wavelet, which has no compact support, we saw in many simulat ions, that
a least square approximat ion can lead to much better results than with well known wavelets with co mpact support. So in that
article we do an error estimat ion for the Shannon wavelet, if we use not all bases coefficients.
Keywords: component; formatting; style; styling; (insert minimu m 5 to 8 key words)

I. INTRODUCTION
In the wavelet theory a scaling function is used, which
has properties that are defined in the MSA (mult i scale
analysis). Through the MSA we know, that we can
construct an orthonormal basis of a closed subspace Vj ,
where Vj belongs to a sequence of subspaces with the
following property:

c
k

j ,k

c
k

j ,k

|k|kmax

2
k

|k|kmax

||.|| is the L2 (R) norm.

Here we get the following basis coefficients:

... V1 V0 V1 ... L (R),

2

(1a)

{ j,k(t)}kZ is an orthonormal basis of Vj with j,k(t) =

2j/2 (2j t - k).

y j (t ) :

y( t )

j ,k

( t )dt

In the case, that y is not quadratic integrable on R and we

need an approximation on the Interval I, we can calculate
an orthogonal projection of 1 I y on Vj and we get the
following basis coefficients:

We use the following appro ximation function:

k max

ck y , j ,k

j ,k (t )

k k min

(1b)
kmax and kmin depend on the approximation interval I =
[t 0 ,tend].

ck 1I y , j ,k

y( t )

j ,k

( t )dt

That leads in general to a wo rse approximation, if j is not

very big (see [7]).
For easier notation we assume k max = - k min .
Now we use the Shannon wavelet and assume for
I = [-a, a], a > 0 and k 0 := k max [2j a] where [x] is the
smallest integer n with n

For k max < we get the following erro r estimation:

340

x. ck = 1I y, j ,k . Additional

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)

we assume, that k max > [2j a] . Then we get the following

error estimation :

In the following examp le we will see, that if we calculate

the coefficients ck by the minimization of

(2)

c
k

j ,k

j ,k

|k | k max

2 j/2c

(2)

|k | k max

k
k 1

Q (c )

(3)

j (t i )

y (t i )

i 0

we can get much better results as if we calculate the

coefficients over (1a) or (1b). In the examp le we use
equidistant points ti (that means ti+1 - ti = t). The method
we call the (discrete) least square method.

Here is the Riemann function and c depends on a and y.

Proof:

ck

y
m

k0

Generally we know that

y(t )

j , k (t ) dt

y(t )

j , k (t ) dt

y y jR

2a max y (t ) j , k (t ) 2a max y (t ) max j , k (t )

t
I
tI

tI

L2 ( I )

if min y g

: c

gV j

y yI j

L2 ( I )

y y jR

Now we consider the last factor:

and

sin( (2 j t k ))

max j ,k (t ) 2 j / 2 max
tI

(2 j t k )

tI

2 j/2

max
tI

1
2jt k

min y g
gV j

L2 ( I )

y y jI

L2 ( I )

So here yj R is the best approximat ion (in reference to the L2

norm) of y on R and yj I is it's best approximat ion on I.

Because of k > 2j a

II. EXAMPLES

2jt k

In all examp les we set j = 1. The appro ximation interval I is

[-1, 1]. We now show three examples, one with a L2 (R)
function, one with a function in V1 and in the last examp le,
we appro ximate a function, wh ich is not in L2 (R) but in
L2 (I). Here we project 1 [-1,1] y in V1 .

has on I = [-a, a] it's maximu m at the point t = a.

So we get

max j ,k (t )

2 j/2

k 2 a

tI

In all examples we co mpare the orthogonal projection fro m

y with the least square estimation (see (3)) with t 0 = -1.

1) y is in L2 (R):

and

ck

y (t ) e t and set k max = 15.

2

2 j/2c

k 2ja

We use the function

a) Orthogonal projection on V1 :
Now we get:
Here is the graph of y (dashed) and yj :

c
k

j ,k

c
k

j ,k

2 j/2c

|k | k max

k k max 1 (k

1.0

2 a)
j

0.9
0.8

If we set k0 =
=

k max 2 a if a is an integer or generally k 0

j

0.7

k max 2 j a , then we get the proposition:

0.6
0.5

c
k

j ,k

c
k

|k | k max

j ,k

2 j/2c

k k 0 1

0.4

1.0

k2
Figure 1.

341

0.5

0.0

0.5

1.0

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)

The graph of y - yj :

b) Least square approximat ion like in (3) described. We set

t = 1/20.

Here is the graph of y (dashed) and yj :

1.0
5. 10 6

0.9
0.8

1.0

0.5

0.5

1.0

0.7
0.6

5. 10 6

0.5
0.4

Figure 2.

1.0

For y and yj on [-3,3]:

Figure 5.
1.0

0.5

0.0

0.5

1.0

The graph of y - yj :
1. 10 10

0.8

5. 10 11

0.6
1.0

0.5

0.5

1.0

5. 10 11

0.4

1. 10 10

0.2
1.5 10 10

2. 10 10

2.5 10 10

Figure 3.
Figure 6.

For y - yj on [-3,3]:
Here we see that the approximat ion on the interval [-1, 1]
can even be used for an ext rapolation:

5. 10 6

For y and yj on [-3,3] (ext rapolation):

1.0

0.8

0.6
5. 10 6

0.4
0.2

Figure 4.

1
0.2

Now we see the approximat ion function calculated through

the least square method (3).

Figure 7.

For y - yj on [-3,3] (ext rapolation):

342

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)
15

0.35

14

0.30

13
12

0.25

11
10

0.20

9
0.15

8
7
6
5
4
3
2
1

0.10
0.05

0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

Figure 8.

Here we see a table of the coefficients calculated like in

(1a) named ckL(R) and the coefficients calculated over the
least square approach named ck.

ckL(R)

k
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1

0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

9.98646 107
1.12346 106
1.27115 106
1.44692 106
1.65714 106
1.90945 106
2.2138 106
2.49637 106
6.39081 106
0.0000837542
0.00136911
0.0129464
0.0745338
0.260124
0.550701
0.707101
0.550701
0.260124
0.0745338
0.0129464
0.00136911
0.0000837542
6.39081 106
2.49637 106
2.2138 106
1.90945 106
1.65714 106
1.44692 106
1.27115 106
1.12346 106
9.98646 107

ck

|(ckL(R) - ck)/ ckL(R)|

0.0549149
0.162391
0.156844
0.0958456
0.11707
0.143967
0.00448466
0.309462
0.0238892
0.227517
0.0770245
0.00674154
0.0744649
0.26013
0.550695
0.707107
0.550695
0.26013
0.0744648
0.00663831
0.0799417
0.247064
0.0756419
0.193832
0.254212
0.282422
0.125028
0.124006
0.0955985
0.0000702266
0.287446

54 988.4
144 544.
123 386.
66 242.3
70 647.
75 396.
2026.77
123 966.
3739.06
2717.49
57.2587
0.479275
0.000924883
0.0000222773
0.0000111669
8.87617 106
0.0000111669
0.0000222773
0.000926093
0.487248
59.3894
2950.86
11 837.
77 646.3
114 832.
147 907.
75 447.2
85 704.4
75 207.3
63.5093
287 835.

ck

ck

|(ck

L( R)

- ck)/ ck

0.0555138
0.00279287
0.0908575
0.0753896
0.182931
0.274183
0.0337608
0.354246
0.0793108
0.250574
0.0757508
0.00727581
0.0744744
0.26013
0.550695
0.707107
0.550695
0.26013
0.0744823
0.0086523
0.0482002
0.115958
0.0431607
0.0832123
0.226748
0.00943628
0.154941
0.186859
0.0550771
0.132146
0.138983

Here we see the errors y y j

for the different

ckL(R):
8.9239210-6
least square ck with t=1/20:
9.0077110-11
least square ck with t=1/100:
6.412210-11

And here the errors for a bigger interval (extrapolat ion)

y yj

L2 ([ 1.5,1.5])

ckL(R):
0.0000107337

L(R)

L2 ([ 1,1])

55 588.1
2486.95
71 475.6
52 102.6
110 391.
143 592.
15 249.2
141 905.
12 411.1
2992.78
56.3283
0.438006
0.000797106
0.0000222771
0.0000111669
8.87617 106
0.0000111669
0.000022277
0.000690303
0.331684
36.2054
1385.51
6752.56
33 334.3
102 426.
4940.88
93 498.1
129 144.
43 327.5
117 623.
139 171.

approximation methods:

We get similar results with a s maller t = 1/100:

L(R)

9.98646 107
1.12346 106
1.27115 106
1.44692 106
1.65714 106
1.90945 106
2.2138 106
2.49637 106
6.39081 106
0.0000837542
0.00136911
0.0129464
0.0745338
0.260124
0.550701
0.707101
0.550701
0.260124
0.0745338
0.0129464
0.00136911
0.0000837542
6.39081 106
2.49637 106
2.2138 106
1.90945 106
1.65714 106
1.44692 106
1.27115 106
1.12346 106
9.98646 107

least square ck with t=1/20:

343

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)

0.0000248848

y(t) and yj(t) on [-3,3]:

least square ck with t=1/100:

1.0

0.0000196081
2) y is in V1 :

0.5

Here we use a part of the orthonormal basis of V1 to

construct a function in V1:

0.5

25

y (t )

j , k (t ) with ck=(-1) k/(k2 +1).

Figure 11.

k 25

For y - yj on [-3,3]:

a) Orthogonal projection on V1 :

0.0003

We set kmax = 15 (so yj y).

0.0002
0.0001

Here is the graph of y (dashed) and yj :

1
0.0001
0.0002
0.0003

1.0

Figure 12.
0.5

b) Least square approximation like in (3) described k max =

10 and t = 1/20:
1.0

0.5

0.5

1.0

Here is the graph of y (dashed) and yj :

0.5

1.0

Figure 9.
0.5

The graph of y - yj :

1.0

0.5

0.5

0.00008
0.00006

Figure 13.

0.00004
0.00002

1.0

0.5

0.5

0.5

1.0

0.00002

The graph of y - yj :

Figure 10.

344

1.0

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)
10.
9.
8.
7.
6.
5.
4.
3.
2.
1.

4. 10 10
3. 10 10
2. 10 10
1. 10 10

1.0

0.5

0.5

1.0

0.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.

1. 10 10
2. 10 10

Figure 14.

For y and yj on [-3,3] (ext rapolation):

0.00990099
0.0121951
0.0153846
0.02
0.027027
0.0384615
0.0588235
0.1
0.2
0.5
1.
0.5
0.2
0.1
0.0588235
0.0384615
0.027027
0.02
0.0153846
0.0121951
0.00990099

4.16328
5.46063
0.84032
0.911112
0.0585332
0.0363541
0.0678428
0.0998917
0.2
0.5
1.
0.5
0.2
0.0998525
0.0746509
0.184991
0.805369
0.396838
0.6829
0.626755
1.40537

419.491
448.772
53.6208
44.5556
3.16573
1.94521
0.153328
0.00108337
7.19026 1011
1.75861 1010
9.0701 1011
1.52477 1010
8.76749 1011
0.00147549
0.269064
5.80977
28.7987
20.8419
45.3885
52.3939
140.943

For a co mparison we set kmax = 15 and t = 1/ 100:

1.0

0.5

k
3

15.
14.
13.
12.
11.
10.
9.
8.
7.
6.
5.
4.
3.
2.

0.5

Figure 15.

For y - yj on [-3,3] (ext rapolation):

0.10

1.

0.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.

0.05

0.05
0.10
0.15
0.20
0.25

Figure 16.

ckL(R)

ck

|(ckL(R) - ck)/ ckL(R)|

345

ckL(R)
0.00442478

0.00507614
0.00588235
0.00689655
0.00819672
0.00990099
0.0121951
0.0153846
0.02
0.027027
0.0384615
0.0588235
0.1
0.2
0.5
1.
0.5
0.2
0.1
0.0588235
0.0384615
0.027027
0.02
0.0153846
0.0121951
0.00990099
0.00819672
0.00689655
0.00588235
0.00507614
0.00442478

ck

|(ckL(R) - ck)/ ckL(R)|

0.171272
0.0487982
0.497604
0.247044
0.117465
0.621908
0.0267993
0.196127
0.205125
0.177672
0.1073
0.0531263
0.10006
0.2
0.5
1.
0.5
0.2
0.100071
0.0515767
0.135855
0.308153
0.318697
0.0523028
0.204147
0.0272149
0.137145
0.0496725
0.218437
0.209121
0.0107344

39.7075
10.6132
85.5926
34.8214
15.3307
61.8127
1.19754
13.7482
9.25624
7.57388
1.78981
0.096853
0.000604441
6.2635 1010
8.38828 1011
4.66002 1011
8.86247 1011
6.92697 1010
0.000711496
0.123196
2.53223
12.4017
14.9348
4.39968
15.74
3.74871
15.7316
6.20252
38.1342
42.1968
3.42598

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)

Here we see the errors

y yj

2. 10 10

L2 ([ 1,1 ])

for the different

1.5 10 10

approximation methods:
1. 10 10
5. 10 11

ck

L( R)

:
1.0

0.5

0.5

0.0000708509

5. 10 11

least square ck with t=1/ 20 k max = 10:

1. 10 10

1.0

1.5 10 10

1.8087510-10
Figure 17.

least square ck with t=1/ 100 k max = 15:

8.2089210-11

For y - yj on [-3,3] (ext rapolation):

0.25

And here the errors for a bigger interval (extrapolat ion)

y yj

L2 ([ 1.5,1.5])

0.20

0.15

ckL(R):

0.10

0.00013277

0.05

least square ck with t=1/ 20 k max = 10:

0.000050488

Figure 18.

least square ck with t=1/ 100 k max = 15:

0.0000257716
3) y is in 2 ([-1,1]):
Here we use the non L2(R) function y(t) = e-t .
For k max = 15 and t = 1/ 100:
a) At first we calculate the orthogonal projection of 1 [-2,2] y
on V1 (like in (1b) described):

The graph of y - yj :

Here is the graph of y (dashed) and yj :

346

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)
Figure 21.
2.5

2.0

1.5

The graph of y - yj :
1.0

0.5
1. 10 9
1.0

0.5

0.5

1.0
1.0

Figure 19.

0.5

0.5

1.0

1. 10 9

2. 10 9

The graph of y - yj :
3. 10 9

0.3

Figure 22.
0.2

0.1

For y and yj on [-3,3] (ext rapolation):

1.0

0.5

0.5

1.0

0.1

20
0.2

15

Figure 20.
10

b) Least square approximation like in (3) described with

kmax = 15 and t = 1/20:

Figure 23.

Here is the graph of y (dashed) and yj :

2.5

For y - yj on [-3,3] (ext rapolation):

2.0

1.5

1.0

0.5
1.0

0.5

0.5

1.0

347

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)
3.0

least square ck with t=1/ 100:

2.5

4.9004510-10

2.0

1.5

And here the errors for a bigger interval (extrapolat ion)

y yj

1.0

L2 ([ 1.5,1.5])

0.5

Figure 24.

ckL( R)

k
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1

0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

0.045956

0.0505131
0.0560689
0.062991
0.0718511
0.0835905
0.0998736
0.123941
0.163037
0.237215
0.428941
2.34877
3.6603
1.6616
1.34176
0.574112
0.536393
0.168756
0.240166
0.0124894
0.0488908
0.0473013
0.0440978
0.0409305
0.0380603
0.0355111
0.0332544
0.0312522
0.0294685
0.0278719
0.0264356

ck

|(ckL( R) - ck)/ ckL(R)|

0.636832
3.9634
0.40329
1.79733
0.709774
1.46529
2.04353
1.32979
5.97515
1.31779
6.26748
5.09408
3.16799
1.92212
1.16582
0.707107
0.428882
0.26013
0.157235
0.0470357
0.492759
1.32103
0.00904417
0.563427
0.99555
1.25702
0.215155
0.117797
0.369592
0.961601
1.18362

Here we see the errors y y j

L2 ([ 1,1])

ckL(R):

14.8574
77.4629
8.19276
29.5331
10.8784
18.5293
21.4612
11.7292
35.649
4.55524
15.6115
1.16883
0.134501
0.156783
0.131127
0.231653
0.200433
0.541458
0.345307
4.76604
11.0788
26.928
1.20509
14.7655
27.1572
36.3981
5.46998
2.76924
11.5419
33.5008
43.7738

0.353685
least square ck with t=1/ 20:
0.00032093
least square ck with t=1/ 100:
0.000178703

In all examp les the least square method (which is a

numerical approach to get the "best" solution on the
approximation interval I) leads to very good results. Even
to better results as with the best L2 approximation on R.

III. CONCLUSION
In all examp les the least square method (which is a
numerical approach to get the "best" solution on the
approximation interval I) leads to very good results. Even
to better results as with the best L2 approximation on R.

for the different

approximation methods:

A. Text Font of Entire Document

The entire document should be in Times New Ro man or
Times font. Other font types may be used if needed for
special purposes. Type 3 fonts should not be used.
Reco mmended font sizes are shown in Table 1.

ckL( R):
0.219725

B. Title and Author Details

Title must be in 20 points Times New Ro man font. Author
name must be in 11 points times new ro man font. Author

least square ck with t=1/ 20:

1.0156810-9
348

COMPUSOFT, An international journal of advanced computer technology, 2 (11), November-2013 (Volume-II, Issue-XI)

affiliation must be in 10 points italic Times new ro man.

Email address must be in 10 points times new ro man font.

F. Table Captions
Tables must be numbered using uppercase Roman
numerals. Table captions must be centred and in 8 pt
Regular font with Small Caps. Every word in a table
caption must be capitalized except for short minor words as
listed in Section III-B. Captions with table numbers must
be placed before their associated tables, as shown in Table

All tit le and author details must be in single-column format

and must be centered. Every wo rd in a title must be
capitalized. Email address is compulsory for the
corresponding author.
C. Section Headings

Sr.
No
.

No more than three levels of headings should be used. All

headings must be in 10pt font. Every word in a heading
must be capitalized except for short minor words as listed
in Section III-B.
Level-1 Heading: A level-1 heading must be in Small
Caps, centered and numbered using uppercase Roman
numerals. For example, see heading III Page Style of this
document. The two level-1 headings which must not be
numbered are Acknowledg ment and References.

Heading
1

Heading
2

Heading
3

Headin
g4

Heading
5

Headin
g6

G. Page Numbers, Headers and Footers

Page numbers, headers and footers must not be used.
H. Links and Bookmarks
All hypertext lin ks and section bookmarks will be removed
fro m papers during the processing of papers for
publication. If you need to refer to an Internet email
address or URL in your paper, you must type out the
address or URL fu lly in Regular font.

Level-2 Heading: A level-2 heading must be in Italic,

left-justified and numbered using an uppercase alphabetic
letter fo llo wed by a period. For examp le, see heading C.
Section Headings above.
Level-3 Heading: A level-3 heading must be indented,
in Italic and numbered with an Arabic numeral followed by
a right parenthesis. The level-3 heading must end with a
colon. The body of the level-3 section immed iately fo llo ws
the level-3 heading in the same paragraph. For examp le,
this paragraph begins with a level-3 heading.

IV. REFERENCES
The heading of the References section must not be
numbered. All reference items must be in 8 pt font. Please
use Regular and Italic styles to distinguish different fields
as shown in the References section. Nu mber the reference
items consecutively in square brackets (e.g. [1]).

D. Figures and Tables

[1] M. Schuchmann, M. Rasguljajew, 2013. "Error Estimation of an

Figures and tables must be centered in the column. Large

figures and tables may span across both columns. Any table
or figure that takes up more than 1 column width must be
positioned either at the top or at the bottom of the page.

Approximation in a Wavelet Collocation Method". Journal of

Applied Computer Science & Mathematics, No. 14 (7) / 2013,
Suceava.
(http://jacs.usv.ro/index.php?pag=showcontent&issue=14&year=201
3).

E. Figure Captions

[2] M. Schuchmann, M. Rasguljajew, 2013. "Error Estimation and

Figures must be numbered using Arabic nu merals. Figure

captions must be in 8 pt Regular font. Captions of a single
line must be centered whereas mult i-line captions must be
justified. Captions with figure nu mbers must be placed
after their associated figures

Assessment of an Approximation in a Wavelet Collocation Method".

American Journal of Computational Mathematics, Vol.3, No.2, June
2013.

[3] M. Schuchmann, M. Rasguljajew, 2013. "An Approximation on a

Compact Interval Calculated with a Wavelet Collocation Method can
Lead to Much Better Results than other Methods". Journal of
Approximation Theory and Applied Mathematics, Vol. 1.

[4] M. Schuchmann, M. Rasguljajew, 2013. "Extrapolation and

Approximation with a Wavelet Collocation Method for ODEs".
Journal of Approximation Theory and Applied Mathematics, Vol. 1.

[5] M. Schuchmann, M. Rasguljajew, 2013. "Determination of Optimal

Parameters in a Wavelet Collocation Method". International Journal
of Emerging T echnology and Advanced Engineering, Vol. 3, Issue
5, May 2013.
http://www.ijetae.com/files/Volume3Issue5/IJETAE_0513_01.pdf

[6] M. Schuchmann, M. Rasguljajew, 2013. "Approximation of Non

L2(R) Functions on a Compact Interval with a Wavelet Base".
Journal of Approximation Theory and Applied Mathematics, Vol. 2.
M. Schuchmann, M. Rasguljajew, 2013. "Implementation and
Testing an Algorithm for a Wavelet Collocation Method in
Mathematica". International Journal of Emerging T echnology and
Advanced Engineering, Vol. 3, Issue 6, June 2013.
http://www.ijetae.com/files/Volume3Issue6/IJETAE_0613_01.pdf

349