6 views

Uploaded by diogorossot_85125119

Método da Partição da Unidade, elementos finitos.

- tamu-cs-tr-2004-7-4
- Mirna Dzamonja and Saharon Shelah- On properties of theories which preclude the existence of universal models
- 713083 2 Interpolation
- PPT on Binomial Theorem_Sec 4 a Math
- How to read a hard math book
- Itgs Calculators
- Gompertz
- 4
- 3RD Term Pointers for Review in Math
- Technical Computing Laboratory Manual
- IJMTT-V9P502
- group5e
- nps34
- MCQ-NA-A
- SMO Algebra I
- soln.docx
- Aleks
- Gprw-closed Set, Sanjay Mishra, Varun Joshi, Nitin Bhardwaj
- miaa360 lesson study 7th grade
- Lagrange Interpolation

You are on page 1of 17

Adriano Scremin

6 de setembro de 2006

Sumario

1 Introduction 2

2 Some preliminaries 2

3 Getting a procedure 5

6 Examples 12

6.1 Diagonal C cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

6.1.1 Case 1: s = 2 and k=1 . . . . . . . . . . . . . . . . . . . . . 13

6.1.2 Case 2: s = 3 and k=1 . . . . . . . . . . . . . . . . . . . . . 13

6.1.3 Case 3: s = 3 and k=2 . . . . . . . . . . . . . . . . . . . . . 14

6.2 Non-diagonal C case . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

A Appendix 15

2 SOME PRELIMINARIES

1 Introduction

Customers of the Generalized Finite Element Method (GFEM), the Partiton of

Unity Method (PUM), the hp-Clouds Method (hp-CM) and the Extended Finite

Element Method (EFEM) have the experience of exactly reproducing polynomials,

under suited conditions, by the usual form:

X s

X s

X

P (x) = i (x) ai + bni cn (x xi ) , x , (1)

i n=k+1 =0

s

X

where {i } forms a k-reducible partition of unity for the set {xi } and cn (xxi )

=0

are s k linearly independent polynomials of degree greater than k and less or equal

s.

Nevertheless, there is no guarantee wether such representation, Eq.1, exists or

not, and under which conditons it holds, except for polynomials of degree less or

equal k, which can be easily verified by making bni = 0, n, and taking into account

{i } is k-reducible. A close clue is given by the Reproduction Formula proposed by

Duarte, but it is still a bit far from evidence the existence of such form, Eq.1.

This article is focused on the following question: are there coefficients ai and bni

in Eq.1 satisfying the identity above?, or, equivalently, is there a procedure to find

them?, and under which hypothesis do they exist?

2 Some preliminaries

Before demonstration and in order of clearness, it should point out some relevant

propositions and theorems.

First consider a k-reducible partition of unity {i } on for the set {xi } as defined

by Duarte. By definition, such partition of unity exactly reproduces polynomials of

degree less or equal k. This was shown by Duarte in . Consequently any polynomial

P of degree less or equal k verifies:

X

P (x) = i P (xi ), x . (2)

i

for n = k + 1, k + 2, , s and = k + 1, k + 2, , s, where is the Kronecker

delta. Matricialy it would look like this:

2

2 SOME PRELIMINARIES

c0(k+1) c0(k+2) c0s

c1(k+1) c1(k+2) c1s

.. .. .. ..

. . . .

ck(k+1) ck(k+2) cks

C = (cn )(s+1)(sk) = (3)

1 0 0

0 1 0

.. .. .. ..

. . . .

0 0 1

Plays a fundamental role as the starting point to achieve the desired procedure

the Reproduction Formula proposed by Duarte:

following formula holds for all polynomials P of degree s 1 :

X s

X

()

P (x) = i (x) P (xi ) + P (xi )(x xi ) , x . (4)

i =k+1

Before going on, let make a pause to introduce an ilustrative problem. At first

glance, the coefficients ai and bni in Eq.1 should be straightly determined comparing

Eqs.1 and 4. But soon one is in touch with a not simple task. Inside the square

brackets in Eq.1, after changing the order of the summation and spliting them,

results for comparison with Eq.4:

X s

X k

X s

X s

X s

X

i ai + bni c0n + (x xi ) bni cn + (x xi ) bni cn .

i n=k+1 =1 n=k+1 =k+1 n=k+1

As can be seen above, the second summations inside the square brackets remain

with no pair of comparison in Eq.4, deserving a tricky manner to overcome it.

Especial and important results will be shown and demonstrated in the following.

The last one gives a tricky clue to find the procedure.

Proposition 2.1 Let {i } be a k-reducible partition of unity on for the set {xi }.

Let m be an integer such that 0 m k. Then:

X

(x xi )km i = 0, x . (5)

i

1

The superscript () in Eq.4 stands for derivative of order , and the coefficients are given

recursively in Duarte.

3

2 SOME PRELIMINARIES

Proof: Developing the left term in the above equation, taking into account the

binomial theorem:

X km

km k m

X X

(x xi ) km

i = (1) x xkm

i i

i i =0

km X

km k m

X

= (1) x xkm

i i . (6)

=0

i

X

xkm

i i = xkm , x .

i

km

X

km

X k m km

km

(x xi ) i = (1) x

i =0

km

km k m

X

km

= x (1) = 0. (7)

=0

complete.

Proposition 2.2 Assume the same hypothesis of Proposition 2.1. Let n be any

positive integer and mn be a real. Then:

n

X

n km

XX n n

mn x (x xi ) i = mn xi (x xi )km+ i = 0, x .

i i =0

(8)

X

mn xn (x xi )km i = 0.

i

4

3 GETTING A PROCEDURE

X

mn xn (x xi )km i =

i

X

= mn (x xi + xi )n (x xi )km i

i

n

X X n

= mn (x xi ) xn

i (x xi )km i

i

=0

n

XX n n

= mn xi (x xi )km+ i .

i =0

3 Getting a procedure

With the help of Proposition 2.2, the Reproduction Formula can be rewritten for

xs as:

X s

X s

s

x = i xsi + ! xs

i (x xi ) +

i =k+1

k1 X

sk X

n

X n n km+

+ mn xi (x xi ) .

m=0 n=0 =0

In the above equation the last term in square brackets is null. This is remarkable

because is the starting trick to achieve the procedure. In order to group both

summations in the square brackets, calling the power of x xi , the power of

xi and adjusting accordingly the extremes in the right sum, after a lot of algebra

which can be found in Appendix A, one get 2 :

2

In what follows, the upper bound lower than the inferior means no summation.

5

3 GETTING A PROCEDURE

X

s

x = i xsi +

i

s

X s

+ ! xs

i (x xi ) +

k+1

sk min(k1,s)

k X

X X ++mk

+ m(++mk) xi (x xi )

=1 =0 m=k

+mk

s s min(k1,s)

X X X ++mk

+ m(++mk) xi (x xi )

=k+1 =0 m=0

+mk

++mk

Am = m(++mk) ,

+mk

r() = min(k 1, s )

and grouping the third and fourth terms in the square brackets in the summation

over , Eq.9 yields:

X

s

x = i xsi +

i

k sk r()

X X X

+ (x xi ) xi Am

=1 =0 m=k

s s r()

X s s

X

X

+ (x xi ) ! xi + xi Am

=k+1

=0 m=0

Using the Kronecker delta, the term on xsi inside the square brackets can be

replaced by a summation of xi terms, resulting:

6

3 GETTING A PROCEDURE

X

s

x = i xsi +

i

k sk r()

X X X

+ (x xi ) xi Am

=1 =0 m=k

s s r()

X

X s X

+ (x xi ) xi ! (s) + Am

=k+1 =0

m=0

X s

X n

X

s

x = i ai + bni cn (x xi ) . (9)

i n=k+1 =0

For the purpose of comparison between both, Eq.9 and Eq.9, dismembering the

right term above gives:

X s

X

s

x = i ai + bni c0n (10)

i n=k+1

k

X s

X

+ (x xi ) bni cn + (11)

=1 n=k+1

Xs Xs

+ (x xi ) bni cn . (12)

=k+1 n=k+1

After subtracting Eq.9 from Eq.10, and remarking that the identity is verified

for any (x xi ) , results three new equations:

s

X

ai = xsi bni c0n , (13)

n=k+1

s sk r()

X X X

bni cn = xi Am , 1 k, (14)

n=k+1 =0 m=k

s s r()

X X s X

bni cn = xi ! (s) + Am , k + 1 s. (15)

n=k+1 =0

m=0

from Eq.15:

7

3 GETTING A PROCEDURE

s r()

X s X

bi = xi ! (s) + Am , k + 1 s. (16)

=0

m=0

s sn r(n)

X X s X

cn xi n! (sn) n + Anm =

n=k+1 =0

n m=0

sk r()

X X

= xi Am , k + 1 n s and 1 k, (17)

=0 m=k

On the left side of the above equation, changing the order of the summations

and extending the summation over till s k with the help of the Kronecker delta,

gives:

sk s r(n)

X X s X

xi (1 (sk) )cn n! (sn) n + Anm =

=0 n=k+1

n m=0

sk r()

X X

= xi Am , k + 1 n s and 1 k, (18)

=0 m=k

s r(n)

X s X

(1 (sk) )cn n! (sn) n + Anm =

n=k+1

n m=0

r()

X

= Am , 0 s k 1. (19)

m=k

Or rearranging in such a manner that the A coefficients are left on the same side:

r() s r(n)

X X X

Am (1 (sk) )cn Anm =

m=k n=k+1 m=0

s

X s

= n! (1 (sk) )(sn) n cn , 0 s k 1. (20)

n=k+1

n

8

3 GETTING A PROCEDURE

nishes:

r()

X ++mk

m(++mk)

m=k

+mk

s r(n)

X X n++mk

(1 (sk) )cn m(n++mk) =

n=k+1 m=0

n+mk

s

X s

= n! (1 (sk) )(sn) n cn , 0 s k 1. (21)

n=k+1

n

r()

X ++mk

m(++mk) =

m=k

+mk

r()

X ++mk

= (k) + m(++mk) ,

0 m=k+1

+mk

+ 1 + + m k s k 1. (22)

s r(n)

X X n++mk

(k) =(1 (sk) ) cn m(n++mk)

n=k+1 m=0

n+mk

r()

X ++mk

m(++mk) +

m=k+1

+mk

s

X s

+(1 (sk) ) n! (sn) n cn ,

n=k+1

n

+ 1 n + + m k s k 1 and

+ 1 + + m k s k 1. (23)

9

4 LESS DEGREE POLYNOMIAL REPRODUCTION

notes that = s k makes the upper bound in the first and second summations less

than the inferior one, i.e., no sum involving coefficients takes place. This means

that coefficients of the form (k)(sk) can be obtained independently of any

coefficients. These two remarks above allow the following algorithm for determining

the coefficients:

begin

for from s k down to 0 do

for from 1 to k do

determine (k) by Eq.23

end

The procedure already developed is restricted to polynomials of degree s. No

effort has been devoted yet to less degree polynomials, i.e., degree between k and s.

The question is wether exist ai and bni such that:

X s

X s

X

p

x = i (x) ai + bni cn (x xi ) , k < p < s. (24)

i n=k+1 =0

To begin with, consider p, k < p < s. It was proved that there are ai and bni

such that:

p p

X X X

p

x (x) = i (x) ai + bni cn (x xi ) , x , (25)

i n=k+1 =0

p

X

noting cn (x xi ) are of order less or equal p, and holding the same hypothesis

=0

on cn , 0 s, Eq.3. With no loss of generality, in Eq.24, remarking cn = n ,

k + 1 n, s, and taking ai = ai , bni = bni , k + 1 n p, and bni = 0,

p + 1 n s, results:

10

5 EXISTENCE AND UNIQUENESS

X s

X s

X

i (x) ai + bni cn (x xi ) =

i n=k+1 =0

p s s s

X X X X X

= i (x) ai + bni cn (x xi ) + bni cn (x xi )

i n=k+1 =0 n=p+1 =0

p s

X X X

= i (x) ai + bni cn (x xi )

i n=k+1 =0

p p p s

X X X X X

= i (x) ai + bni cn (x xi ) + bni n (x xi )

i n=k+1 =0 n=k+1 =p+1

p p

X X X

= i (x) ai + bni cn (x xi ) = xp .

i n=k+1 =0

Once the coefficients are found, coming back to Eq.16, bi , k + 1 s,

can finally be obtained. Consequently, ai can be calculated with the aid of the

bi already computed. Moreover, as the above algorithm obviously shows, the

coefficients exist for all s and {xi } chosen for the partition of unit, providing the

hypothesis on cn be held. So do bi and ai .

The hypothesis on cn may even be relaxed, say sub-matrix (cn )sk(sk) must be

invertible, which is guaranteed by the hypothesis on the linear independence of the

Xs

polynomials cn (x xi ) . To see this, let cn , 0 s and k + 1 n s, be

=0

s

X

coefficients such that there exist c1

n verifying cm c1

mn = n , k + 1 , n s.

m=k+1

Taking:

s

X

cn = cm c1

mn , 0 s, (26)

m=k+1

it has already been shown that there exists bni such that:

X s

X s

X

p

x (x) = i (x) ai + bni cn (x xi ) , k + 1 p s. (27)

i n=k+1 =0

s

X

Assuming ai = ai , bni = cnl bli and considering Eq.26, after substituting in

l=k+1

11

6 EXAMPLES

Eq.27, results:

X s

X s

X s

X s

X

p 1

x = i ai + cnl bli cm cmn (x xi )

i n=k+1 l=k+1 =0 m=k+1

X s

X s

X s

X s

X

= i ai + bli cnl cm c1

mn (x

xi )

i =0 l=k+1 n=k+1 m=k+1

X Xs Xs Xs

= i ai + bli ml cm (x xi )

i =0 l=k+1 m=k+1

X Xs Xs

= i ai + bmi cm (x xi )

i =0 m=k+1

X Xs Xs

= i ai + bmi cm (x xi ) ,

i m=k+1 =0

s

X

So, there exist ai = ai and bni = c1

nm bmi such that:

m=k+1

X s

X s

X s

X

xp = i ai + bni

cn (x xi ) . (28)

i =0 n=k+1 =0

s

X

trix (cn )(sk)(sk) is invertible, or equivalently, whenever the polynomials cn (x

=0

xi ) , k + 1 n s, are able to generate polynomials of the form:

The uniqueness of such coefficients is beyond the scope of this article, and is

left for future work. But the point is wether fi 0 is the unique solution of

X

i (x)fi (x) = 0, x , or not.

i

6 Examples

To ilustrate the theorem just presented, consider two series of examples. The

first refers to the especial case which the coefficientes c are of the type of Eq.3, and

the second

12

6.1 Diagonal C cases 6 EXAMPLES

6.1.1 Case 1: s = 2 and k = 1

c02

C = c12

1

ai = c02 + x2i

bi = { b0i 1 }T

X

i (2x2 x + c12 xi c12 x) =

i

X X

= i (2x2 x) = i x2 = x2

i i

c02 c03

c12 c13

C =

1 0

0 1

ai = 3c02 (c12 + xi ) + 2c03 + x3i

bi = { b0i 3(c12 + xi ) 2 }T

X

i (3xi x2 2x3 + 3c212 xi 3c212 x + 3c12 xi x 3c12 x2 + 2c13 xi 2c13 x) =

i

X X

= i (3x3 2x3 ) = i x3 = x3

i i

13

6.2 Non-diagonal C case 6 EXAMPLES

c03

c13

C =

c23

1

ai = c03 + x3i

bi = { b0i b1i 1 }T

X

i (x3 + 3x2i x 3xi x2 + c23 x2 + c23 x2i 2c23 xi x + c13 x c13 xi ) =

i

X X

= i (x3 + 3x3 3x3 ) = i x3 = x3

i i

1 1 1

1 1 1

C=

1 2 1

1 1 2

2 1 1

1 1 1

4 4 4

1 1 1

4 4 4

1 0 0

0 1 0

0 0 1

29 11 3

ai = + xi + x2i + x4i

16 4 2

b1i

19 + 11 x + 3 x2

16 4 i 2 i

bi = 15 1 9 2

16

x

4 i

x

2 i

5 21 x + 3 x2

16 4 i 2 i

14

A APPENDIX

X 29 29 11 11 2 3 2 3 3

i 4xi x 3x + xi x + xi x x + xi x x = x4

3 4

i

16 16 4 4 2 2

A Appendix

First remark the restrictions on m, n and :

0mk1 (29)

0nsk1 (30)

0 n. (31)

By definition, let:

=km+ (32)

= n . (33)

+ = n m + k. (34)

one immediatelly finds:

1 s 1.

Eqs.29-30 and Eq.34 to 0 s 1. Joining both, results the restriction on

:

min(s 1, s k 1) = s 1, if > k. This fact suggests dismembering

the summation on in two others summations.

Moreover:

15

A APPENDIX

0mk1

0n=++mk sk1k ms1

0 m + k n k m n + k = m + ,

or equivalently:

0 m min(k 1, s 1).

Taking into account the results above one can rewrite:

k1 sk1

X X X n X X min(k1,s1)

k sk1 X X X min(k1,s1)

s1 s1 X

= + .

m=0 n=0 =0 =1 =0 m=k =k+1 =0 m=0

16

Referencias REFERENCIAS

Referencias

17

- tamu-cs-tr-2004-7-4Uploaded byesteban_quito_21
- Mirna Dzamonja and Saharon Shelah- On properties of theories which preclude the existence of universal modelsUploaded byHmsdk
- 713083 2 InterpolationUploaded byRho Natta Matahari
- PPT on Binomial Theorem_Sec 4 a MathUploaded bynadia sykes
- How to read a hard math bookUploaded byJaqui Cabañitas
- Itgs CalculatorsUploaded byReza Tandisau Juventini
- GompertzUploaded byaquamariaxtine5914
- 4Uploaded byLuis Fuentes
- 3RD Term Pointers for Review in MathUploaded byEnzo Alcantara
- Technical Computing Laboratory ManualUploaded byKarlo Untalan
- IJMTT-V9P502Uploaded byHussein Alkafaji
- group5eUploaded byapi-240269666
- nps34Uploaded byAkshay Viswanathan
- MCQ-NA-AUploaded byvrsafe
- SMO Algebra IUploaded byJmex Mk
- soln.docxUploaded byAyush Gupta
- AleksUploaded bywendy
- Gprw-closed Set, Sanjay Mishra, Varun Joshi, Nitin BhardwajUploaded byvarunjoshi_1985
- miaa360 lesson study 7th gradeUploaded byapi-245618218
- Lagrange InterpolationUploaded byKuhu Koyaliya
- JEE_Class 11_SPUploaded byAditya
- Chapter 2Uploaded byjayar0824
- Buch BergerUploaded byAntonio Madueño Luna
- On some locally closed sets and spaces in Ideal Topological SpacesUploaded byIJMER
- BA Logic 6Uploaded byIkki De Queero
- Wolfram Bentz- Optimal Packings of 13 and 46 Unit Squares in a SquareUploaded byMnaom
- Alg 2 H Syllabus '14Uploaded byMichelle New
- Is science the supreme form of knowledge?Uploaded bygennaghy
- Algebra Basics Fundamentals - Elementary.docxUploaded byNeil Isaac Perez
- Practica 3Uploaded byScrib_3

- UM BREVE ESTUDO SOBRE AS ESTRUTURAS TENSEGRITY.pdfUploaded bydiogorossot_85125119
- Domos GéodesicosUploaded bydiogorossot_85125119
- Montagem de Domo de 2m de DiâmetroUploaded bydiogorossot_85125119
- Exercicios resolvidosUploaded byBia Pimentel
- 0221060_05_cap_03Uploaded byDaniel Marques Silva
- Professional Dome PlansUploaded bychrysto_gypsy
- Dynamics of Structure Chopra 1995Uploaded bylmc_12036596
- UM BREVE ESTUDO SOBRE AS ESTRUTURAS TENSEGRITY.pdfUploaded bydiogorossot_85125119
- Apostila de FortranUploaded bydiogorossot_85125119
- Trabalho de programação avançada de computadores.pdfUploaded bydiogorossot_85125119
- Mechanical Vibration solved examplesUploaded byParas Thakur

- C1 2006 May 2006 Mark SchemeUploaded byHashan Godakanda
- solution seç. 10.4 Diprima e boyceUploaded byharyly
- Lehmann IA SSM Ch9Uploaded byAnonymous 7gpjR0W
- Peculiar modules for 4-ended tanglesUploaded byFábio Duarte
- January 2007R QP - C4 EdexcelUploaded byShwan Hasan
- NCERT Solutions Class 12 Maths Chapter 11 Three Dimensional GeometryUploaded byVidyakul
- Representation of Linear Groups copy.pdfUploaded bymayk
- Introduction to Mathematics of General RelativityUploaded byimanarrovi
- Dual SpacesUploaded byAim Libra
- Introduction to Abstract Algebra [Lecture Notes]Uploaded byJohn
- Information Technology Course Description2Uploaded byRomeo Balingao
- CAPE Pure Mathematics 2017 U1 P2Uploaded byYagna Lall
- b 03308011Uploaded byInternational Journal of computational Engineering research (IJCER)
- Calculator Techniques PERCDCUploaded bySharelyn Nebreja
- Chapter3.2 Linear Algebraic Equation - LU in & Matrix InversionUploaded byAhmad Fadzlan
- Trigonometric IdentitiesUploaded byMikhail López
- GabeUploaded byKaram Aneja
- BACAMAR1Uploaded byVA GA MA
- SyllabusUploaded byMOHIT SINGH
- 65(B) MATHS FOR BLIND CANDIDATES.pdfUploaded bySabaresh
- Tutorial PARI GPUploaded byEmmanureva
- HammingCodes.pptUploaded byFrancis Sachin
- Multivariate Tutsheet 1Uploaded byabhishek010594
- Parameterizing Curves SurfacesUploaded byAdnaan Mukadam
- 7 a Curves and SurfacesUploaded byumeshsharma123
- Dimensionality ReductionUploaded byghromis
- dscsc.pdfUploaded bymssms
- Chapter 2 Quadratic Equations Paper 1 With AnswerUploaded byJarnice Ling Yee Ching
- benaichUploaded byhamidf
- 4037_w10_ms_13Uploaded bynaseeb