&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
Å
,QWHUSRODomR
2 3UREOHPD Dado um conjunto de pontos,
[ \ [ \ « [Q \Q com [L z [M para L z M , com L M Q
determinar uma IXQomRGHLQWHUSRODomR
M[ L
M tal que,
\L L «Q
SRUH[HPSOR
Dado o conjunto de pontos,
duas possíveis soluções seriam,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
7HUPLQRORJLD
x
Os valores [ [ [Q chamam-se QyVGHLQWHUSRODomR e
x
O conjunto ^ [L \L L Q`chama-se VXSRUWHGHLQWHUSRODomR
x
^
x
Existem váriosWLSRVGHIXQo}HVGHLQWHUSRODomR, tais como:
os respectivos \ \ \Q são os YDORUHVQRGDLV
M[ L
\L L «Q` é a IXQomRGHLQWHUSRODomR nesse suporte x
,QWHUSRODomRSROLQRPLDO
x
,QWHUSRODomRWULJRQRPpWULFD
onde
x
0 é um inteiro igual a Q se Q é SDU e Q se Q é tPSDU,
L
é a unidade imaginária
,QWHUSRODomRUDFLRQDO
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Å
3ROLQyPLRV
GHILQLomR
Um SROLQyPLRGHJUDX Q é uma função da forma,
onde DQ z , excepto se Q D D DQ os FRHILFLHQWHV do polinómio
Å
2 HVTXHPD GH +RUQHU
Como FDOFXODURYDORU de um polinómio num dado ponto?
H[HPSOR
p3(x) = a3 x3 + a2 x2 + a1 x + a0
{ Q Q
p3(x) = (( a3 x + a2 ) x + a1 ) x + a0
{ Q = PXOWLSOLFDo}HV }
DOJRULWPR
= QQ = PXOWLSOLFDo}HV }
{ Entrada: a0, a1, ..., an , x ∈ ¸ }
polinomio ← an
para k desde n ± 1 até 0 fazer
polinomio ← polinomio ∗ x + ak
n
n-1
{ Saída: polinomio = an x + an-1 x
FRPSOH[LGDGH
+ ... + a1 x + a0 }
Q multiplicações e Q adições
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
WHRUHPD ( +RUQHU )
Para calcular
pn(c), YDORUGHXPSROLQyPLR de grau n QRSRQWR c,
faça-se
b n = an
e calcule-se
bk = ak + c bk+1
então
b0 = pn(c)
para k = n-1, n-2, ...,1, 0
q0(x) = bn xn-1 + bn-1 xn-2 + ... + b2 x + b1
Mais ainda, se
então
pn(x) = (x - c) q0 + r0
onde
q0(x) é o SROLQyPLRFRFLHQWH de grau n-1 e
r0 = b0 = pn(c) é o UHVWR.
H[HPSOR Calcular
p5() = x5 - 6 x4 + 8 x3 + 8 x2 + 4 x - 40 pelo esquema de Horner,
p5() =
b5 = a5
=1
b4 = a4 + 3 b5
=-6+3
=-3
b3 = a3 + 3 b4
= 8-9
=-1
b2 = a2 + 3 b3
= 8-3
=5
b1 = a1 + 3 b2
= 4 + 15
= 19
b0 = a0 + 3 b1
= - 40 + 57
= 17
ficando assim também calculado o SROLQyPLRFRFLHQWH,
T[ = bn xn-1 + bn-1 xn-2 + ... + b2 x + b1 = x4 - 3 x3 - x2 + 5 x + 19
e o UHVWR,
U = E = 17
de modo que:
S[ = [FT U = (x - 3) (x4 - 3 x3 - x2 + 5 x + 19 ) + 17
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Assim, os resultados parciais obtidos pelo PpWRGRGH+RUQHU são efectivamente os
sucessivos valores calculados pela 5HJUDGH5XIILQL :
D
D
D
D
D
D
1
-6
8
8
4
-40
3
-9
-3
15
57
-3
-1
5
19
[ 1
E
E
E
E
E
= S
E
permitindo calcular:
p5(x) = (x - 3) (x4 - 3 x3 - x2 + 5 x + 19 ) + 17
,QWHUSRODomR 3ROLQRPLDO
Å
Os SROLQyPLRV são excelentes candidatos a IXQo}HVLQWHUSRODGRUDV, porque:
x
x
x
x
O FiOFXORGRVYDORUHV é realizável em RUGHPOLQHDU no número de
multiplicações e adições.
As operações de GHULYDomRHSULPLWLYDomR são simples e podem ser facilmente
programáveis.
São de classe &
ˆ
$SUR[LPDPWDQWRTXDQWRVHTXHLUDqualquer IXQomRFRQWtQXD num LQWHUYDOR
ILQLWR.( Teorema de Weierstrass ).
Sempre que as funções de interpolação consideradas são polinómios falamos em
,QWHUSRODomR3ROLQRPLDO.
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
2 3UREOHPD Dado um VXSRUWHGHLQWHUSRODomR com Q pontos,
^[L \L L «Q `
encontrar um SROLQyPLRGHJUDXd Q tal que,
\L
4XHVW}HV
x
x
SQ [L L «Q
([LVWH sempre um polinómio que satisfaz as condições acima?
Caso exista, é ~QLFR?
7HRUHPDGDH[LVWrQFLDHXQLFLGDGH
Seja 3Q o conjunto dos polinómios de JUDXPHQRURXLJXDOD Q.
Dados Q SRQWRVVXSRUWHGLVWLQWRV [L \L L «Q,
H[LVWHXPHXPVySROLQyPLR SQ  3Q tal que,
SQ [L \L L «Q
'HPRQVWUDomR
Seja
A exigência de que SQ seja um SROLQyPLRLQWHUSRODGRU nos Q pontos [L \L traduz-se no VLVWHPD de equações lineares:
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Para que o sistema tenha VROXomR~QLFD é QHFHVViULRHVXILFLHQWH que a respectiva
matriz dos coeficientes possua um GHWHUPLQDQWHGLIHUHQWHGH]HUR.
A PDWUL]GRVFRHILFLHQWHV é a conhecida PDWUL]GH9DQGHUPRQGH, definida por:
Comecemos por demonstrar que o GHWHUPLQDQWHGH9DQGHUPRQGH tem o valor,
Atendendo a que [L z [M para L z M (SRQWRVGLVWLQWRV) , este determinante
é QmRQXOR e portanto o sistema tem VROXomR~QLFD.
'HPRQVWUDomR do YDORUGRGHWHUPLQDQWHGH9DQGHUPRQGH:
por LQGXomR sobre Q :
x Para Q , verifica-se pois,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
x Suponha-se a propriedade válida para Q í e mostre-se que é válida para Q
Multiplicando a SULPHLUDFROXQD por [ e VXEWUDLQGR o resultado à VHJXQGDFROXQD,
multiplicando a VHJXQGDFROXQD por [ e VXEWUDLQGR o resultado à WHUFHLUDFROXQD, ...
obtemos,
desenvolvendo este determinante e factorizando ( [ ± [ da primeira linha, [ ± [
da segunda linha, ..., [Q ± [ da última linha ) obtemos,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
assim, temos:
Aplicando agora a KLSyWHVHGHLQGXomR ao determinante de Vandermonde de ordem Q
e multiplicando, obtemos para a ordem Q :
Portanto, se é esta a H[SUHVVmRGRGHWHUPLQDQWH da PDWUL]GRVLVWHPD e, atendendo a
que [L z [M para L z M , este determinante é QmRQXOR então o sistema tem VROXomR
~QLFD.
2EVHUYDo}HV
x
Existem GXDVGHVYDQWDJHQV que não tornam recomendável computacionalmente
seguir esta via para GHWHUPLQDURSROLQyPLRLQWHUSRODGRU:
A matriz de Vandermonde apresenta um Q~PHURGHFRQGLomR muito
elevado, tanto maior quanto maior for Q, pelo que se trata de um SUREOHPD
PDOFRQGLFLRQDGR.
x
Trata-se de um processo de cálculo SRXFRHILFLHQWH – é possível obter o
polinómio interpolador com menos operações aritméticas.
O teorema anterior mostra-nos que o polinómio interpolador H[LVWHHp~QLFR
(deduziremos YiULDVIyUPXODV para ele, mas todas representam R PHVPR
SROLQyPLRLQWHUSRODGRU).
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
>
3ROLQyPLRV GH /DJUDQJH
GHILQLomR
Os polinómios de grau Q dados por,
são designados por SROLQyPLRVGH/DJUDQJH
DVVRFLDGRVDRVQyV [ [ [Q
WHRUHPD
O SROLQyPLRLQWHUSRODGRU SQ de grau menor ou igual a Q que interpola os
valores nodais \ \ \Q nos nós distintos [ [ [Q é dado por,
GHPRQVWUDomR
Pela sua definição, os polinómios /N satisfazem a relação,
onde
Gkj é o delta de Kronecker
nestas condições,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Assim, este polinómio cujo grau é evidente ser menor ou igual a Q, interpola os valores
dados. Pelo Teorema da existência e unicidade é também o único polinómio interpolador
nestes pontos.
H[HPSOR
Construir o polinómio interpolador de grau ≤ 3 que interpola os seguintes valores:
Os SROLQyPLRVGH/DJUDQJH associados
aos nós ( x0 = 0 , x1 = 1, x2 = 3 , x3 = 4 )
obtêm-se directamente da GHILQLomR,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Assim sendo, nas condições do teorema, o SROLQyPLRLQWHUSRODGRU é dado por:
2EVHUYDomR
A fórmula de Lagrange SRGHQmRVHUDUHSUHVHQWDomRPDLVFRQYHQLHQWH do polinómio
interpolador. Isto acontece, fundamentalmente por duas razões:
É possível obter este polinómio com PHQRVRSHUDo}HVDULWPpWLFDV que as
requeridas por aquela fórmula;
Os polinómios de Lagrange HVWmRDVVRFLDGRVDXPFRQMXQWRGHQyV e uma
mudança de posição ou do número destes altera completamente estes
polinómios.
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
>
)yUPXOD GH 1HZWRQ
GHILQLomR
.
A )RUPDGH1HZWRQ para polinómios de grau Q é dada por,
onde os parâmetros FL , L Q são chamados FHQWURVGRSROLQyPLR
&RQVWUXomR da )yUPXOD GH 1HZWRQ:
Considerando os QyV [ [ [Q como FHQWURVGRSROLQyPLR, temos:
Os FRHILFLHQWHV D D DQ vão ser determinados de modo a que SQ seja o
SROLQyPLRLQWHUSRODGRU nos nós [ [ [Q dos valores nodais \ \ \Q :
ou, se os valores nodais \L forem os YDORUHVQRGDLVGHXPDIXQomR I temos,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Assim, a partir de,
e fazendo sucessivamente x = x0 , x = x1 , ... , x = xn obtemos os FRHILFLHQWHV:
REVHUYDomR
CadaFRHILFLHQWH
x
x
DN N
Q:
pode ser calculado a partir dos DL L N, Mi GHWHUPLQDGRV.
GHSHQGH exclusivamente dos QyV [ [ [N e dos respectivos
YDORUHVQRGDLV \ \ \N
GLIHUHQoDGLYLGLGDGHRUGHPN ( N • ) entre os N nós [ [ [N .
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Å
'LIHUHQoDV 'LYLGLGDV
Para designar a GLIHUHQoDGLYLGLGDGHRUGHPN ( N • ) entre os N nós [ [ [N,
são utilizadas indistintamente GXDVQRWDo}HV:
onde,
ou
WHRUHPD
Os FRHILFLHQWHV DN N Q do polinómio
SQ
de grau d Q ,
na IRUPDGH1HZWRQ que interpola os valores I[ , I[ , ..., I[N
nos QyVGLVWLQWRV [ [ [N
são dados LQGXWLYDPHQWH pela expressão:
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Assim, o3ROLQyPLR,QWHUSRODGRUFRP'LIHUHQoDV'LYLGLGDV tem a forma:
&iOFXORGDV'LIHUHQoDV'LYLGLGDV
As diferenças divididas são calculadas por FRQVWUXomRGHXPDWDEHOD,
tal como no seguinte caso para 4 nós:
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
H[HPSOR
Determinar o SROLQyPLRLQWHUSRODGRU, na IRUPDGH1HZWRQ,
que interpola os seguintes pontos,
Tabela de GLIHUHQoDVGLYLGLGDV:
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Assim, calculados os FRHILFLHQWHV do polinómio interpolador na IRUPDGH1HZWRQ,
temos,
REVHUYDo}HV
x
A RUGHP pela qual os nós são tomados é DUELWUiULD.
x
Se os valores nodais forem os valores nodais de uma IXQomR, é possível
estabelecer uma OLJDomR importante entre as GLIHUHQoDVGLYLGLGDV de ordem N e a
GHULYDGD da mesma ordem dessa função.
x
Se fosse necessário DFUHVFHQWDUPDLVDOJXPQy aos anteriores, bastaria colocálo no fundo da tabela e calcular mais uma linha de valores (as diferenças divididas
anteriormente obtidas não seriam afectadas).
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
UHFRUGHPRV
7HRUHPDGR9DORU0pGLR/DJUDQJH
Para toda função contínua em [a, b] e derivável em (a, b)
existe (pelo menos) um ponto F ∈ (a, b) onde:
&DVRSDUWLFXODU±7HRUHPDGH5ROOH
Se
então
&DVRSDUWLFXODUGRWHRUHPDGH5ROOH
Se
então
Entre GRLV ]HURVGDIXQomR existe (pelo menos) XP ]HURGDGHULYDGD.
&RUROiULRGRWHRUHPDGH5ROOH
Entre Q ]HURVGDIXQomR existem
(pelo menos) Q ]HURVGDGHULYDGD.
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
WHRUHPD Sejam I ∈ & ( [a, b] ) e [ [ [Q QyVGLVWLQWRV no intervalo [a, b].
Q
Então existe um
[ ∈ (a, b) tal que,
GHPRQVWUDomR
Para Q , o Teorema do Valor Médio garante esse resultado,
Para analisar o caso geral Q !, consideremos a função,
onde SQ[ é o polinómio de grau d Q que LQWHUSROD a função nos Q nós distintos.
Assim, HQ[tem (pelo menos) Q zeros distintos em (a, b) e, pelo Corolário do
Teorema de Rolle,
tem (pelo menos) Q zeros em (a, b)
aplicando sucessivamente o Corolário do Teorema de Rolle,
tem (pelo menos) Q zeros em (a, b)
tem (pelo menos) zero em (a, b)
e seja
[ esse zero.
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Por outro lado, derivando Q vezes a expressão do Polinómio Interpolador com Diferenças
Divididas,
obtemos,
Portanto,
ou,
x
Deste modo, se os valores nodais forem os valores nodais de uma função, este
teorema estabelece uma UHODomRLPSRUWDQWH entre as GLIHUHQoDVGLYLGLGDV de
ordem Q e a GHULYDGD da mesma ordem dessa função.
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
>
,QWHUSRODomR FRP 1yV (TXLGLVWDQWHV
Em muitas aplicações os QyV [ [ [Q são HTXLGLVWDQWHV.
Sendo K a GLVWkQFLD entre dois nós sucessivos ( DYDQoR ou SDVVR) temos,
donde,
Å
'LIHUHQoDV 3URJUHVVLYDV RX 'HVFHQGHQWHV GHILQLomR
A GLIHUHQoDSURJUHVVLYD de RUGHP]HUR e SDVVR K da função I no nó [L
é dada por,
A GLIHUHQoDSURJUHVVLYD de RUGHPN , N e SDVVR K
da função I no nó [L é dada por,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Tal como as diferenças divididas, as diferenças progressivas organizam-se numa WDEHOD:
Além disso, existe uma relação entre as diferenças divididas e as diferenças progressivas.
WHRUHPD
A GLIHUHQoDGLYLGLGDde ordem N da função I nos N nós equidistantes de passo K
[L [L [LN está UHODFLRQDGD com a GLIHUHQoDSURJUHVVLYD de ordem N por,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
recordemos,
H GHPRQVWUHPRV
SRULQGXomR
Para N ,
por definição
e para N ,
Suponhamos agora a relação válida para N Qt e provemos para N Q :
a partir da relação,
e aplicando a hipótese de indução,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
obtemos,
)yUPXOD GH 1HZWRQ FRP 'LIHUHQoDV 3URJUHVVLYDV
WHRUHPD
O SROLQyPLR de grau ” Qque LQWHUSROD os valores nodais I I IQ
nos QyVHTXLGLVWDQWHV [ [ [Q de passo K
pode escrever-se na forma,
GHPRQVWUDomR
Aplicar o resultado anterior à Fórmula de Newton com Diferenças Divididas,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Å
'LIHUHQoDV 5HJUHVVLYDV RX $VFHQGHQWHV GHILQLomR
A GLIHUHQoDUHJUHVVLYD de RUGHP]HUR e SDVVR K da função I no nó [L
é dada por,
A GLIHUHQoDUHJUHVVLYD de RUGHPN , N e SDVVR K
da função I no nó [L é dada por,
Analogamente, existe uma relação entre as diferenças divididas e as diferenças
regressivas.
WHRUHPD
A GLIHUHQoDGLYLGLGDde ordem N da função I nos N nós equidistantes de passo K
[LN [LN [L está UHODFLRQDGD com a GLIHUHQoDUHJUHVVLYD de ordem N por,
GHPRQVWUDomR
Por indução sobre N.
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Também a GLIHUHQoDVUHJUHVVLYDV se organizam numa WDEHOD:
)yUPXOD GH 1HZWRQ FRP 'LIHUHQoDV 5HJUHVVLYDV
WHRUHPD
O SROLQyPLR de grau ” Qque LQWHUSROD os valores nodais I I IQ
nos QyVHTXLGLVWDQWHV [ [ [Q de passo K
pode escrever-se na forma,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
GHPRQVWUDomR
Basta considerar a Fórmula de Newton com Diferenças Divididas, mas
construída relativamente aos nós por RUGHPLQYHUVD [Q [Qí [ [ ,
e aplicar o resultado anterior.
>
(UURV GH ,QWHUSRODomR 3ROLQRPLDO
Que HUUR se comete quando seLQWHUSRODXPDIXQomR por um SROLQyPLR de grau d Q
utilizando o valor da função em Q nós distintos
"
SRUH[HPSOR
I[ VLQ[[
S[
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
WHRUHPD Sejam I 
&Q([a, b]) e SQ
o SROLQyPLR de grau d Q
que LQWHUSROD I nos QyVGLVWLQWRV [ [ [Q, contidos em [a, b] .
Então para TXDOTXHU
∈ [a, b] H[LVWH um valor
dependente de
[
∈ (a, b) ,
e de I tal que
GHPRQVWUDomR
x
x
Nos QyV da interpolação,
o HUUR é igual a ]HUR e o resultado é verdadeiro.
Para analisar os RXWURV pontos,
consideremos o produto,
e a função auxiliar,
)[ tem pelo menos Q zeros em [a, b] que são:
então, aplicando sucessivamente o Corolário do Teorema de Rolle,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
)[ tem pelo menos Q zeros em [a, b]
)¶[ tem pelo menos Q zeros em [a, b]
)¶¶[ tem pelo menos Q zeros em [a, b]
)Q[
tem pelo menos zero em [a, b]. Seja
[ um desses zeros.
Assim,
com,
e porque,
temos
e portanto,
Esta expressão calcula o valor exacto do HUURGHLQWHUSRODomR em qualquer ponto
VH o valor de
[ for conhecido.
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
QRWD
x
Para [  >@ a função [ ±[ tem o seu valor máximo em [
Assim, nesse intervalo,
x
.
[ ±[d Para [  >DE@ com K
E±D , analisemos a função [±DE±[ :
D
[
E
Fazendo [ DVK , com V  >@ , ao valor máximo de V corresponde [ DK donde, no intervalo >DE@ ,
[±DE±[d K x
ou
_[±D[±E_d K Para Q pontos equidistantes [ [ [Q com passo K , analisemos,
e provemos que
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
por LQGXomR sobre Q .
•
Para Q , como vimos, _ [±[ [±[ _d K •
Assumindo que, para
temos,
provemos que para
teremos
Efectivamente,
[
[
[
[Q
[Q
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
Å
(VWLPDWLYD GR (UUR GH ,QWHUSRODomR
Como em,
o valor de [ é desconhecido, temos de calcular um OLPLWHVXSHULRUpara estimativa do
valor do erro.
Para o caso particular da função a interpolar, procuramos um PDMRUDQWH em [x0, xn],
e considerando K o HVSDoDPHQWRPi[LPR entre dois nós consecutivos,
temos,
ou,
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
&RPSRUWDPHQWR GR (UUR GH ,QWHUSRODomR
Å
Analisando,
verificamos que o erro de interpolação depende de:
x
o Q~PHURGHQyV considerado,
x
o comportamento do SROLQyPLR Z de grau Q .
x
o comportamento da GHULYDGDGHRUGHP Q da função,
O comportamento do SROLQyPLR Z pode ter efeitos indesejáveis, tal como no conhecido
caso do (IHLWRGH5XQJH :
A IXQomRGH5XQJHé definida em [ -1, 1] por,
QyV HTXLGLVWDQWHV
QyV HTXLGLVWDQWHV
QyV HTXLGLVWDQWHV
SROLQyPLRVLQWHUSRODGRUHV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
x
Verifica-se que, FRQVLGHUDQGRQyVHTXLGLVWDQWHV, os polinómios interpoladores
tendem a RVFLODUQRVH[WUHPRV e que oscilam WDQWRPDLV quanto PDLRUIRUR
Q~PHUR de nós considerado!
x
Para QyVHTXLGLVWDQWHV com [ [ [Q verifica-se que R SROLQyPLR
RVFLOD muito nos LQWHUYDORVH[WUHPRV >[ [@ e >[Q [Q@ e menos nos intervalos
centrais.
x
Z
Prova-se que o valor
é PtQLPR ( e igual a
x
íQ) quando os nós coincidem com os QyVGH&KHE\VKHY.
No intervalo >í@ os QyVGH&KHE\VKHY são:
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
&DStWXOR±$SUR[LPDomRGH)XQo}HVHGH'DGRV
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
x
Num intervalo >DE@ os QyVGH&KHE\VKHYsão dados por:
onde WN N Q representa os nós de Chebyshev em >í@ .
x
Usando os nós de Chebyshev verifica-se que o SROLQyPLRQRGDO exibe RVFLODo}HV
XQLIRUPHV em todo o intervalo considerado.
x
&RPHVWDGLVWULEXLomRHVSDFLDOGRVQyV é possível mostrar que, se I for uma
função contínua e diferenciável em >DE@, o SROLQyPLRLQWHUSRODGRUFRQYHUJH
para I quando Q o
ˆ
para todo o [ >DE@
BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
$QiOLVH1XPpULFD5RViOLD5RGULJXHVH$QWyQLR3HUHLUD
Download

CapítuIo 3 ± Aproximação de Funções e de Dados