One-Step Method .

· Euler's Method [Taylor's Method of order 1 ,

Forward

1). Definition :

Mesh points :

ti =

atih Vi = 0 1, 2 , N in the interval [a b]

need
, . . ., ,

Because InLtn+a)
v
no
Step Size :

hi =

Fits-t:
always >0 ! know
to
[Explicit Mtd]
.

Let Yuctn) be an approximation for Yctn) Sit .

YuCtn+a) Yuctn) +hofctn

=

, YnLtnil
with Exact solution :

Yctn)

Differential equation with Euler's Method Y(tu) fath ynctnl) f(tn yn)
: =

=
. ,

2) Psedocode :

3) Error Analysis :

Since Yctn+h) By Taylor Expansion

Yctn+1) =

<
=Yctn) + h Y(tn)
.
+bY"(3n) for 3nECtn ,
tn+)

Hence Y(tn) =
-Yctn) + och) , Y"(3n) truncation error .

Y
Note that fith YnCtrl)=,

Thus the error for fith Ynctnl) Differential equation with Euler Method Och)
=

For Euler's Approximation :

Yutz= Ynth fcth Yn) ·

=
Yn+Och
i e
. . Zochy for each step of approximation .

After n
steps ,
the
accumulating error :
n .

och") :

nh och)
-
:
(tn-to) och)
·

① Therefore ,
the total error
:

Och) .

i e . . Ycth)-Yactull = B h .

for some BEIR depending on th .

② .
General bound for the error :

Let k =

max1 41 ,
M-
maxlY" its
Since Yatate) Yctn) +h Yctn) =

. + Y" (3n) for some 3nt (tn ,

tn+1)

y(tn+1) =

yctn) h yctn) +by"(3n)

+
·

for some Ent (tn , tnta)

3n)
So ente =

enth Sty entEh2Y"3n) .

Intel Itn) 21 +
h k) + Ihi
-

Thus , Itz1 1901 11thk) + IhiM

lea = (e) (1thk) + IhM

i e.
=(80)(1thk) + (1 + (1 + h() I hiM
:
11thk) ] IhiM
-

lea Hol Ith"

+21thk)"+
+ .

...

UK
e 1 V + r2+
+

...
+ rut =
I
Therefore ,
Ienl 10/21thk" + A thM .

eirc (1tr)"
Note that ,
inequality ,

lenl = (e) 21thks" +

1
.
EhM
k

i e
. .

-
Itol(1+(n-tolk)"+ EK- 1
. thM
Yo Y
marly
=
=
to =

P .

=>
lenk .
In

③ Asymptotic
.
Error
Analysis :

en= Yn-Yn= h Dn+Och2 ·

With DCt) s .
t . D'Ct) g(t) D(t)
=
- +

EY"Ct) ,
Dito) =

/y Yet
Y)
g(t)
=

=
)

This is to get a better approximation :

(By Richardson's Extrapolation)

Yctn) YnCEn) h. Dctn)+Och

-
=

Yctn) Yeuctn) 2h DCtn) + Och'

-
·
=

=>

YCEn) -

[2YnCtn) -yenctn)] =
Och
·
The backward Euler's Method & The Trapezoidal Method

1) .
Definition :
Recall Forward Euler Approximation fith Ynctnl)
,
=

Anyctn)
Backward At-h)
Approximation
:

Y(t) - h

sit .
fit ,
Yet) = AYCt-h)
Because its should be known . i e
. . Yct) = YCt-h) +
hofct Yctx) ,

X
[Implicit Mrd] i e
. . Yctit1) ~Y(ti) th fctits Ycti+1)) ·

, , Yo= Yo

<Same as Forward] With Error Analysis :

Mn-Ynl=Ochs ,
/Yit1 -

Yitzl= Och

2)
Absolutely Stable Method :

After n
steps ?
(1) .
Definition
:

Y'It xYt) =

,
X 0 lim
Yn im Yuctn)= mytotuh)
= =

0 h
- n
I
-

If Y fit =
,
y)8 .

(2) .

Examples :

⑦ Forward .
Euler's Method NOT
absolutely stable

Proof :

Yitz Yith fiti yi)

= ·

,
=

Yi+hXY : =
(1thX) Yi , Yo= 1 .

Hence ,
yn= (thX) Yn-1 =

AthX)Yn-z =
. ..
=
(1t hX)"

If mn= Em Ithx)" 0
=

Then 11thN1< 1 ) - 1s1thX < 1 Th .

i e .
.
-2hx<0
Since X so ,
then och >
This shows
ImYhtu) 0 is NOT for all h
=

So .
Forward Euler Method is NOT absolutely stable .

② .
Backward Euler Method Absolutely Stable

Proof :

Yitz= Yith fctite ·

,
Yits)
=

Yi th xYi+1= (1-hx) + yi
·

,
Yo= 1 .

= - "

Yn= (1-hx) 21 hx)

-

Yn (1 hx)
1
Yn z
= -
= -
-
=
-
...

If = m1-hx) =
En =

Then 11- hx) > 1

Since XCO ,
any step size >0
So
any h satisfies
Therefore ,
the backward Euler Method is
Absolutely Stable .
 (This is to change implicit method to explicit method)

3) Fixed Points for Backward Euler's Method :

predictor formula
.

① Definition
.
:

Gcy) Yith fctite Yitz)

=
·

,
with first guess
:

Y1= Yith futi ·

,
yil

Then yet =

Yithofctite y ,1) j ,
=
0 1 2,
, ,
...

GCys
Y= GLY Y
Recall NA1 .
fixed -

point iterations =

Yi )
Yi +h ·
f (ti+1 ,

=Yith fctitz Yith fiti Will

F1
· ·

. ,
, ,

imgcy11: Yitz
Y
:

Ih < 1
Yi= gyts(
If I / <M Ft y , =
Yi+h ·
fcti+1 yes ,

=Yith ·

fctitz Yith f Ctitz Yi th fiti Yill)

,
·
, ,

Then hi n
Explicit ?
,
NOW

4. The Trapezoidal Method :

(1) .
Definition :

h
Recall the trapezoidal quadrature rule :
a
-
b

Safixdxfa Paxidx Sat =

fasdx+ fal fab dx a

Ifas + f(b))

With Error :

f(x) Paxs ↑ (x-al(X-b)

=
+

Safexs -Pq(x)dx= x)/(x-a)(x-d)dX= f(x)

=> Yeti+1) Ycti) -
=

St**isds = [Yti)+Y'ti+1)) -
is )"

(Here Y'(s) =

f(x) -

P2())

=
[fiti , till + fatits Y(tie))] ,
-

" 25i)

Hence Yit =Yitz[futi , yil+fctite Yitzl] , ,

Go Yo
=

Implicit With Error =

Och")

(2) .
The total error for Trapezoidal Method :

(After n
steps)

n .
Och= nh Och= (En-to) Och" -
·
=> It's of second order accuracy .

23) .

Trapezoidal Method is Absolutely Stable ·

(Yo Yo
= :

1)

Proof :
Yitz
=

Yi+ [fcti , yil+fcti+1 Yi+11] ,

yi
+

Exyi ExYi+1 +

E) (1 -

Ex)Yit= LI+ ExY : <)

Yitz = y
 Hence .
Yn=("Yo= (EE)" S

= mI =

0 -
/ I
i e
. .
-1 < 1

Since 230 ,
x0 .
So 1-Exs0
-> -
-

1 +
Ex 1 +
Ex<1-EX ,
which always holds for all h.

Therefore trapezoidal ,
method is always absolutely stable .

14) .
Implicit to Explicit :
By Fixed -
Point Iteration .
Henn's Method .

g(y) yi +
=

[fcti yi) + f(ti , +1 , y))

with h must
satisfy :

1) myxy)-1 .

Y
*
= >
=

Yit [futi yi)+fctit1 Y , ,

with first guess :

= Yi+hifcti ,
yi)

y =

Yit[futi yil+fcti+1 yi11] , ,

=yi+ Ifuti yi)+fctita yithf (ti Yill]

,
, ,

Not
Absolutely Stable

Proof :
Yitz =

Yi + :

[fcti yil+f(titz
, , Yitz Yith futi
=
·
,
fill]
yi E xyi x(yi+ h fcti yi))
=

+ .
+ .
.
,

Yi+Ehxyi+Ehxyi+EChx)"y
=

=21 +
hx + Echx)"] Y :

Thus , Un :
[Ithx + EchNsY
"

= Yn = hm[ 1 thx + chx]" 0

=

11+ hx + EchX") < 1

i e . .
- 1<1thx+ = ChN2 < 1
-
<hx +
=24x2 0

i e . .
-< x(1+ Ehx) < O

As ho ,
XCO ,
this implies It Ehx >0
Hence ,
-

ICEXCO -

2 ChXCO
So , Ihx1<2 => h >
1
This shows Henn's Mod is NOT absolutely stable .
·

Taylor Method :

Recall Forward Enter Method is Tayor Method of order 1 .

From Taylor Expansion Yetit1) Y(ti) +h Yctil + 1 Y"ctil+

: =

·
.

...

where forward Euler Method just uses the first two terms of Taylor Expansion .

Yut) = -
Yt) + 2 cost , Yo =

1 .
Exact Solution : Y :

sint+cost

1) .
Example of Taylor Method of order 2 :

From Taylor Expansion :

Yctitz)= Yeti) + h ·
Ycti) + Y"ti) +Och's

[Y(4)] =

Y" (t) = -

Y(t)-2sint=Yct-2cost-2sint
Then , Yitz Yith 2-yi+zwsti) +
=
· .

Lyi-20st: -2sinti) with yo

:
1 .

Hence the at each Ochi) Yi+1 Yi+1

error
step is -

, .

After n
steps ,
nochc :

nh .
Och =

(An-to) Och's .

Therefore ,
the total error is ochi .

2). Example of Taylor Method of order 3 :

From Taylor Expansion

:

Yct+z) =
Yeti) +
h Y cti)
·
+ .

Ycti) + Y"<ti) .
+ Och")

[Y" <til]' =

Y +i) [Y(ti) 2c0st-2sint]'

= -
=

Y'ti) +2 sint-2cost=-Ycti) + 2 Sint

Then ,
Yetits)= Yetil +h ·
( Yti) + 200sti)
-
+ (Ytil-2costi-2sinti) + .
(- Yeti) + 2sinti)

Hence ,
the total error after n
steps
:

Och" .

3). General Formula for Taylor Method of order n :

(1) .
Differentiation on variable t :
+y) =
3)
.
Y (t) =

fit ,
Yct1)
Yt)
Y"(t) =

(t
Yc+1) ,
+
·

Yct) Chain Rule

ft (t , Yt1) +
fyCt , YC) Y'(t) ·

AYt1)
t ,

fit Ct Yct1)
-2nd-order
Est-order
ft)y It Yet) YCt Note & Sty fyt
Y (t)]' +L
=

Ct [Y
"

dt
·
=
= , , .

=[ftCt , Yrt) +
fyct , Yct1) YCt3 ·

-
fit YAI ,

=fit It Yt1) + fty Ct YCH) Y (t) fit Y) + fyy It Y [fet YA11]

>
+

fry ct Y)
·
·

, , ·

,
, , ,

+
fyCt YAI) ftCt YA1) ,
·

,
+ fyct , YA) ·

fyct YA) Y'ct) ,

·

=[ftt + 2 fty f + fyy f+ fy (ft+fyf)] it YA1)· :

·

(2) Taylor Method of order p accuracy :

Yitz
=
Yith-yi+ y, + ... +By with yo: Yo

Error=OchP of order
+
At Each Step ,
) Yp-YP Since yitz p

So Asymptotic analysis Yet-yact)=hP DCt)+och' /

*
error
:

The total error after n

steps : noch" = nh -
Och" (tn-to) Och") =
·
=

Och"/
By Extrapolation :
Yet ynct) hP Dct) + Och+1)
· = .

Yet Y2hct) <2 h) Dct) + OchP+1)

-
P = .

=>
Yct =

2 1(2"yncH -yzhct) + OchPt)

YC-YnCt= Yurt -

YnLt) -

# Y2hLt)
22P 2P
21 Y2n(t)
-
+
1
-1 Yurt)
-

= YnLt)-Yzh(t)
( Y(t) yart)+ 25Lynct) -Yen(i) If Y not known
=
.
·

Runge- Kurta Method :

Same
accuracy as Taylor's ,
but less computation .

1) Definition
.

of Runge :
Kutta Method of e2 :

(with Error Ochill

Yitz =

Yith FCti Yi ·

, ,
h) +Och sit .

Y-Yn= Och
where Fct y h)
, ,
=

Difct y),
+ b2 ·

fct + 2h ,
y Bh f(t y)
+ ·

① Asymptotic
.
error
:

Yet -Yn2t= Dit) h2+ h

-

② . Constant Setting
:

By Taylor Expansion f(x-h)+h) fex-h) +h f/cx-h) +

=
·

fix-h) + ....

With Yet -yets =

och") at each step .

2). Definition of Runge-Kutta Method of Order n

:
=
Error at each step =

Ochrf)

Explicit Method & NOT Absolutely Stable

3) . Tutorial 4 & Implicit Method

.
.
·
Multistep Methods :

Yetiff) Ycti) + Stiff

Yadt
Main idea :
=

. Adams
1 -
Bashforth Method :

EXPLICIT ?

1) .

Definition : Yitz =
Yi+E B fiti . .
,
yi) -fCti-1 Yi-xl] , Interpolation of degree 1 .

Two-step A-B Med .

Proof
let father-actwind
:

Let two points ti-e titit ,

he

Hence ,
P1Ct) =

(p02) K2ti-1)· + L1qCt) K2ti)

=

[Cti-t) K(ti-1) + (t-t=1)- K2til)

-

With the error ,

K2-P1Ct=Ect-tielt-tilk" (3i) By Interpolation Def .

/Pactidt 1 [3kcti)
*

So .
kcdt- =
-
Kuti-e

2) Error
.

Analysis :
St katdt=1Pactdt+ Eh K "(3i) ->
Error at each step :
Och"

Proof :

Since IKLH-Pacts) ECt-Ei) (t-tia) K "Bil

=
.

↳ h

>
I I ·
titl ti ti+1
en

Hence IS kitidt- Pactidt) Ikats -

Pacts/dt

>
Itch2dt=c .

h =

och

Note that , YI is approximate The total error after n steps Ochi.

:

⑦ (Yctn) yn)= (A40ch) + 1Y=-y)

-
Och'+och"= Ochi) =

2 Ye . is estimated by the Euler Method :

Ye Yoth Y:
:
.

So Ma-y1= och'=> Ochi+Ochi =

Och'

4) Definition of Two-Step A-B Method :

Interpolation of degree 2 .
With tiz ,
tia ,
ti

Yitz =

Yi +Ezh [23f(ti yi)-1bfcti-1 y =-1) + 5 fctiz Yi-z)

·

, , ,
Need to know Yo Y1 Ye , , .

Proof :

Let k2t :

f (t Yct) Y(t)
,
= =
P2Ct) =
(2017) ·

f(tiz) + (2=(t) fcti-2) + (22 (t) fiti)· ·

(t-ti -1) (t-ti) (t-ti-z) (t-ti-1)

L20(t) Fitis)
(ti- Ei)
:

,
L21(t) titts
=E ,
Lazt= -

(ti-ti-z) (ti-ti-1)

It /
* **
ktdt Pactsdt =
[23k(ti) -16(s(ti 1) 5kti 2)] -
+

-
2) Error
.
Analysis of Three Step A-B :
(Except initial values)
the

** kctdt= I Pctdt+5 hYk" (3) -> Error at each step Och") :

Proof :
Kuts Parti
-
=

= k"(3i) :

It-tiz) (t-ti-z) (t-ti) By Interpolation def .

h h h
I 1 1 => It- ticksh .
It-tickch , It-tilth
ti ti-I ti tit 1

So / Stidt /Pectsdt1= /kat) - -

Pacts/dtstite-ti) Och =
Och")

The total error after n-2

steps :
(n-2) Och") ·
=

(n-2) h Och"c= (tn-tz)

-
·
·
Och'= Och")

There are
many ways to estimate ya Yz
, .

By Second-order Runge-Kutta Method ,

/Y-yz)=Ochs M2-Yz)= Och' ,

Therefore ,
the total error after n
steps
:

Och'+Och') + Och'l= Och's

3). General Adams -

Bashforth Method :

For a (9+1)-Step A-B method using Interpolation Polynomial of degree

, a with + 1 points

The error at each step

:

Ochaty) .

The total error after n

steps
:

Ochat) =>
Asymptotic error :
Y -YhCt)= Dc h &+ Ochat·

By extrapolation to accelerate
convergence .
2 . Adams -
Moulton Method :
IMPLICITI
1) .

One-Step Adams -
Moulton Method :

Let kCt) =

fit Yct)
,
=

Yct) with points ti ,

titz .

/
*

Recall Main Idea Yetits) Ycti)

:

=
+

ct , YCt)dt

So . (Etidtfbactidt=(stite-t ·
kitil+ (t-til ·

Kutiteldt= [futi Ytill + futite Yctital)

, ,

Therefore ,
Yitz =
YitE [fcti yis+fctive Yixas] , , Implicit .

i e
. .

Trapezoidal rule (Absolutely stable)

2) Error .

Analysis :

Since k2t) P1Ct) + E k (Bi)Ct-ti)(t-tite)

=
"

By Interpolation of degree 1 .

Iti t-rich
t-titzh
,

So ket) Pact)-
= E h h
. .
=

Och2

Then IS Kuts -
Pactsdt1 /ket) > -

Pects/dt= Och At each step .

Mct) Yucts) -
=

(n-1) :

och)+M1-y11 Here y= by Euler's Med .

=(n-1)h Och" · +
/Y= -y=)
=
(An-t1) Och" + Och"
=Och
The total error ,
here , Ochi .

after n
steps .

SF putdt=/
= *

pactdt-22hk"(i)

1) .

Two-Step Adams Moulton Method - :

With points ti-1 ,

ti ,
titz .

Interpolating Polynomial Pact) .

Implicit .

Therefore Yitz yi+ *>[5 fctiff yi+1) + 8 f (ti yi)-f(ti-1 Yi-1l]

. :
=
, , , ,

2). Error Analysis

:

Similar to Adams Bashforth Method -

.

⑦ . At each step
:

Error=Och" .

② Total
. error , after n
steps
=

Och .

S* S
**

KCt) At= Pactdt+h4k 5i)

 predictor formula
y *1 Yi = +
h fiti yi)
·

3). Implicit Explicit

,
->
:

For Two-Step Adams -

Moulton :

PjyI= yith fitite yes , , y*: With fitite yes ,

. . . . .

=> P1 =

Yi +h f (tit1 ·

, Y! (
=Yi +h ·

fctite With fcti yil)

,
·

This is
by support of theory :

1)=12(Yi+2[5ftitz Yitz) 8 frti yi)-fcti 1

Yi)/<1 VE y
+

,
, ,

, .

2y

If hi
maxIy7
.
3 Comparisons of Adams -

Bashforth & Adams -

Moulton :

(h/k "(3i)))
① Second Order
. AM's error (Ehlk"(Bill) - Second Order AB's error

Similar for nith order AM's error - AB's error .

② .
AM :

Implicit ,
AB :

Explicit .

③ .
AM :
Absolutely stable .
AB NOT :

Proof :

Consider two-step AB method ,

Ditz =

Yi +
Eh[3f(ti yi)-fctie , , yi-zl)
Since Y'ct) fct y) x yct)
:

,
=

with x<0
So Yitz Yi+Eh[3 xycti) -xycti-Al]
=
.

(Eh-3x+ 1) yi Eh xYi-1
= - .

A difference equation Yitz (I+Ehx) Yi+Ehxyi-1 0

:
=
-

↳ E>

~S t .
r-2 + Ehx)r+

Un= C1 Vo+Cz Va
·
Ehx

.
=

E)
with roots
nee

(r-VoSCV-V)
Vo

=
,
VI

v=cro+V)U+roV
.

nimYn= m ro+(V) 0 =
IrokI IVK1
=
-

So IEhNc1 i e . .
IhIX , > h >
conflicts Uh .
4. General Multistep Method :

1) .
General Two-Step Multistep Method :

Yitz Yi +As Yiz th [D- fCtit Yitz) +Do fcti yi)+D1 fctin Yi-zl]
SEE
=

do . .
.

, ,
.

⑦ .
Adams -

Bashforth :

Yitz
=

Yith [Efiti
.
,
yi) -Efcti-z , Yi-11]

② Adams Moulton Yitz Yith [ futies Yit) + fiti yi) -F2fct-1 yi-zl]
=
- ·
:

, ,
.