This preview shows pages 1–9. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Numerical Differentiation llO Finite Difference Differentiation 0 h — 0
Formal Deﬁnition of the Derivative f’ (:30) = W If we simply let h be “small”, we have an approximation to the derivative, fa) z _f<wo + h; — f<wo> If h > 0 this is a forward—diﬂerence formula and
if h < 0 this is a backward—diﬁerence formula. It has a unique relation to the Taylor series for f (3:0 + h) mo + h) = mg) + h f’(m0) + gm Mg) , g 6 ($0, $0 + h) f($0+h) _f(x0) mo) = MT— + f”(€) Our approximation to the derivative is obtained by dropping the last term
Which introduces an error of 0(h). The smaller the value of h, the smaller the
mathematical error. However, very small h results in numerical subtraction of two “nearly identical” numbers so it introduces round—off error. A centered difference formula for the derivative is: mg) % f(:ro + h>2—hf<aso — h) The error in this formula is 0(h2) lll “61:13:35 Sometimes the points are not equally spaced so numerical implementation of
the centered difference formula is impossible. Consider the case of y = f
with values at speciﬁc points known as sketched below: <———><——>
L1 l L2 I For all interior points (1132 to 3:6) in the ﬁgure, interpolation of derivatives at the
center of adjacent points can be used to generate the equivalent of a centered
difference formula at each of the X—points, even when the distnces, L, are not
equal. The resulting formula is: / qm _ qm—l 1 [ym+1 ym—l]
f ( m) y 2 gm gm—l 2 gm qm—l ( )
1
where: qm : §(Lm+1 + Lm) For estiating the derivatives at the end points, extapolation can be used from
the numerical derivative half way between the two endmost points using the forward or backward difference formula and the the derivative at the nearest
interior point given by the above formula. 112 I»»/ v.3; “51¢? I Sometimes values of a function, y, are given at unequally spaced points around
the periphery of a plane curve as sketched below. L4 The lengths L are are lengths (s) between points on the curve and values of
y are known at points ( 1, 2, ..., N) on the curve. To obtain the numerical
approximation of the tangential derivative 3/ (s) at points 1,2,3, N, equation
(1) can be used. However, for point m = 1, special values for some of the y’s
and some of the L’s must be used. In particular ym_1 : yN and Lm_1 = LN. Likewise for for point m : N, ym+1 : yl and Lm+1 : L1. 113 UI4 To estimate the error in Simpson’s rule, the function over four successive points
can be expanded in a Taylor series up to order 3. With equally spaced points,
the error in the integral from the cubic term vanishes and the dominant term
in the error is proportional to the fourth derivative d4 f / dart“. For equally spaced
points with h 2 A33, the total error, ET takes the form: M n1 d4f(m) E =——
T 901:1 (1:34 772 is some value of a: in the ith interval. 114 H mama "an.pH .LH ..c/mm.mx mm.mx mm.mx..ﬁvacTLqm
um_mm_:x m..:omQETm new rmuTONQOLF .Lmrzmzmuuwm eoLm mFmmeuCH...c/ mx..wvmp:anm
u O.m £ﬁmH” WH . "new
”m~+xV>+ mﬂ+¥V>%o.v+ nxv> + mH u mH “Nucumna n x Low “AA:V>%m.o +Aﬁv>*m.o + uHV ; u pH unnﬁv>+LHv%; u LH “ucm
"Axvx + pH u.uH
unxv> + LH u LH
Hucnm u x Low
“0 H WH‘
“ouuH
no u LH
umﬂvN  mmvN u g
uvcw
“ma”.w&_ Hnwmvmcmumw u nxv>
and .+&. owmvmcmumw u anN
.cna n x Low
“ma..cx..Uvamcmumw u :
un_L.._vacmnom n upw um.m... "mew: erw pang? qucm.vu:Q:r n pw$
mFZHOm Dmu<mm >44<30m IFHE ZOHF<MUmFZH 4<UHmmZDZ 00 OF FQHMUm&
ucT53c 115 H mama uwmQTLm cam.©
mem.©
nem.m
VOm.m
00m.o
omv.m
onv.m
ovv.o
oov.o
0mm.m
OmN.m
ooa.w
oom.m
00m.m
ooo.m
ooo.v
ooo.o OmOmOmOmOmOmOmOmO
ONMNONmNONLﬁNONmNO
HOOOOHHHHNNNNmmmmv l\ 116 3.5 2.5 1.5 0.5 1.17 Ed mmnvanm
H wamm oow>©.mm pxu.ummmﬂgm AA
mhmvm.mN ooom>.NN mwasm m.comQEHm Ucm HmUHonQmHB \Hmaswcmuomm Eouw mamuwwucH "mama maﬂw uSQCH Hmucm
pcﬂasc AA NOON ~om HmQOpUO zovcﬂz UQmEEoo mmqeﬂz 118 ...
View Full
Document
 Spring '07
 HenrikSchmidt

Click to edit the document details