Error Control and 12kt

V have

E we

that

ti

I y It wi I L E

Ch tech

But be
potentially

c can

large

run

We would like to ensure

that the error in approximation

is less than E

Given of

of

that at each

Imagine
time we have nth
step

order method

thofti yHi h

ylti.ie yHi

1 Och


Which produces

Witi Wit hd Cti wi h

suppose that we use another

method with local truncation

Ch

error

HilthoTCti yHith

ycti.li

t Of h

we assume

E Wit

E Wi
yet

Yltitln Ylti

Fit.ch

fAi.yHihh

yHiti

Wih Wi h

dCti

YHiti withd finish

ha

ycti.ci wit

In similar
way we
get

wit
Fit

1h1 YHiti

corresponding

t of

combining we obtain

Finch Iie.ch tht witi with

och 0cm

since the Left hand side

must

is och we

dominant term

is Titi
Witth

Iit

by

since
assumption

is Och

We to

are use

going
method

our higher order

the

to approximate error

last tine

in the Step

of our lower order method

e define

x
htlwiti

W.tl

Ri

R Ek h

Which

If

the

we
change step

size to 9h then

Ekcehlkke

Titi h

ki
ith f

e R

Thus to the local

keep

than

error less E we

require at

self

IR

ea.hu

il

the

At each
step
calculated
quantity 12

is

and based its size

on

the size used

step

is
accepted and the

algorithm proceeds or

is
rejected modified and

the
step size repeated

If 12 E then

If REE then I
I

and

the is
step size

accepted

h too much

Decreasing

is also hot desirble

we also have a bound on

C I

9 970.1

o.ch

control

We can use error

with different methods

wit

12hL 1 12h5

will also have some

price

to pay


I1
evaluations

12h4 4

12h5 G evaluation

6 4 6 evaluation per
iteration

RKF combines 12h4 12h5

12h5 or Ser

Iia Wit k
191 Kst
ff K

Ks t k

k 137.4kg

wit Wit Ifk LIII

I ke

I see coefficients ki Ko

in the book

with need

Rkf we
only

G evaluations so
comparing

12h45 get decrees

we

of hot in the number of

evaluations

Example

A step one method to solve

if

consistent and only

Ivp is

if I ft.w.co fff

Sol The truncation error

ith
step

ti
YHil.nl

l Tichl olHi yi.id

Then

lusion Tilhl lim

h 10
Mil ycti.it

lo ti Yi h ti
y

i i

OI Hi I Yi co fCti i
yi i IoHiyyii.o

since we assume of is continuous

limnaxk 471 0

h 10

fcti i.fi

have
using la

we

El ti l Yi no


For te a.br wi
yet

and tale

h and i such

that ti ti i at Ci ith

Then fct.hr oIct.w o

we have Cianci h o

h2

G and therefor

Using
limmaxlT.ch

l
oh

ioieiEn
definition

which is
by

consistency

Example 2

Given the modified Euler

method

ithz ffti.hr tfAith withfHi.w

µiw

Let f be continuous and

fry

satisfies a
Lipschitz condition

on 9,63 42 in the variable Y

Show that the modified Eater

method is stable

proof We need to show that

loft w h satisfies a
Lipschitz

condition in W

ztfftcw tIfAth.wthfH4

H W.hr

V ct.W.h eDx a 6Jx o ho

lol t.w.hl octiw.hr l

IIfH.w1 fH.w tfCtth WthfH.w

I

F e

fifth thf Ct E H

thfH.eu
E ltzlw

E lw w

hfft.w EL1w w I

th El

i w Lt Hail

so 01 is
Lipschitz continuous


in W on D

what is the
Lipschitz

constant for

So we showed stability for

the modified Euler method

section 14
step multi step

method

Ivp problems

fct yet

K ae te b

4cal

So far we developed
approximation methods for

single variable single steps

Taylor's method of

Euler order

12h Kkf

With single step explicit methods we use information only from the previous steps
 t In
From m
step multi step methods
we will use information from
the previous n
steps
Definition
for
The m
step multi
step solving Iup problem
has a difference etuation finding
at the
wit mesh
point titi
represented by the following
equation
Wit Am I wi t Cen 2 Wi i t Aowiti m

1 HE bmf Him Wit Ibn i f Hi wi

t too f Titi n wit m
 Ao Ce Um I too bm

when bn o then the method

called otherwise implicit
is
explicit
Wo L Wit Li WELL Wn tidal

Implicit methods are considered

better and allow to
they use
a
larger step size

Example Explicit Us Implicit

Explicit Euler method

yftthl yctlthfft.mx
Implicit Euler method

yctthl yltlthfct.ly tthD

Given y's
Ivy Ky
1 101 1
With implicit method
we
compute
the derivative
at the end of the step size
 th
yctthl yctlthfft.pl
yH1
kyHth1YHth1
Yet
Th
1 Licit

1
explicit
Euler

our
goal derive a

closed form formula for

an m
step method
Recall the fundamental theorem
of calculus

yields 11161 yea

 For ti tie

yin fit y'Hdt

In our case on the

intervals ti j f ti j wi

for ji 0,1 n m I
is not available
fit
we approximate f Ct yet

by intepolating polynomial
pet Lagrange
interpolating polynomial
using the m available
nodes
 tie

fit et w w flutist
ti