=e ee a
654
‘iene a Daeanen Cais 20
[NUMERICAL DIFFERENTIATION.
{nto Eq, (23.2) to yield
fo
i 2k
Soi) — flsss2) —2flnsid + fod
£0) = +0?)
‘or, by collecting terms,
Hebis2) +4 fess) = 3f0)
< +00e) es)
fis) =
Notice that inclusion of the second-derivaive teem has improved the accuracy to
(0{H2), Similar improved versions can be developed forthe backward and centered fore
las as wel as for the approximations of the higher derivatives. The formulas are sumac
rized in Figs. 23.1 through 23.3 along with all the results from Chap. 4. The following
example illustrates the utility ofthese formulas for estimating derivatives.
FIGURE 23.1
Forward fntedvideddierence formulas: wo versions ore preserted for each derivative, The
later version incorperates more tems of fe Taylor series expansion and is, consequently, mare
occur.
Fit Dave Err
va fd) on
fy) = =a Aa 3 ch
second Dacia
p= Masel 2 an
bale siiei+ 20) a)
Tid Deve
Pid ah
rel a)
Foxth Dia
tal = feel on)
Dias) + Mibu) = 2a
+ ad
aa)[cancers |. tame! ‘a.Nanwicl T onawucun-m
Matotsic Engine, Dileniaionand Dotan Conon 250
So ation even
23.1_HIGH-ACCURACY DIFFERENTIATION FORMULAS. 655
Fier Denane Ere
aw
= Melati
frag = Mt ate te) ow)
Sscord Daiaive
ry) = Hel = 2h ad ea
= Bal 2 ad
Bad Sg Mae) op
hid Dae
= Bile) + hac ~ He a
FIGURE 23.2 .
Backward fitecvides ow)
difrence formulas: tuo
versions or prsenied for each
eivotve The lata version fi) + Rec oa
Incorporaies mera ems ofthe
Taylor sais exponsion ond i 262) 2a) + Wea) ed 5
consequenly, mare orci, F ae
FIGURE 23.3, Fist Devatne for
Cantred firitedivided inate .
dlfrerce formats: bao fg — Ht fact or
‘versions ere presered foreach
erative. The lane version fig Redd Meal ote
incorporates more ems othe
“yo sates expansion ond s, Second Detvatve
consequeny, mare accurate fil = 2d + he) on
ful 2d ed :
hay = wf +16 ow
hid Daioh
fsa] — Ds) + ff) = Aca 4
{| Pheu) «i= fea oe
1314 Bile) = Mh + fie) = Bf + fel
oF
ove
39x) + 12R a) + fad
ol)=e ee a
Hethts ergs iro nd Daeanen aan 20
Sx tion terse
656 [NUMERICAL DIFFERENTIATION.
EXAMPIEDSIN HighAccuracy Differentiation Formulas
23.2
Problem Statement, Recall that in Example 4.4 we estimated the derivative of
fe) = 0.x —0.15e — 05x? 0.25" + 1.2
atx = 0.5 using finite divided differences and a step size of h = 0.25,
Forward Backward Centered
th) 1h) oun)
Esinote 1155 ~o718 ~0.934
esl 265 a7 -24
"where the errors were computed on the basis ofthe tre value of ~0.9125, Repeat this com-
putation, but employ the high-accuracy formulas from Figs. 23.1 through 23.3,
Solution. The data needed for this ample is
foi-2) =12
“f0-1) = 11035156
fess) = 0925
Xian “fi1) = 016363281
Mia Sains) =02
‘he oar eet acy Osco 281
7035
‘The backward difference of accuracy O(/2) is computed as (Fig. 23.2)
0935) A.1098186 +12
3035
‘The centered difference of accuracy O(/*) is computed as (Fig. 23.3)
pay a2 4 AOGUER “SLIT 12 yyy
As expected, the errors for the forward and backward differences are considerably
‘more aceurate than the results from Example 44, However, surprisingly, the centered di
ference yields a perfect result. This is because the formulas based on the Taylor series are
equivalent to passing polynomials through the data points.
F105) 0.859375 5.82%
F105) 0878125 = 3.77%
RICHARDSON EXTRAPOLATION
To this point, we have seen that there ate {wo ways to improve derivative estimates when
‘employing finite divided differences: (1) decrease the step size or (2) use a higher-order
formula that employs more points. A thied approach, based on Richardson extrapolation,
uses two derivative estimates fo compute a thind, more accurate approximation,