Modules Interpolation

RootFinding Methods

Numerial Integration Differentiation

Reference Elementary Numerical Analysis CarldeBoor Corte

Some basic result Suppose f ab R is continuous

m n sit f attains its
Max min at these points flat

find MY

If re MM then not a b s t f no 8

fig are cont and takes same signover a b

integrable
g
then ZtCaib Sit
fa.gln.dn ffzlfglal.tn
Rstheorem f.cab R is continuous in ab diff
p
flat flb then it a b sit
file 0
f is cont in ab diff in ab

IE a b St fill ftp.faf

TaylorSeries Expansion i f is Intil times continuously diff

in ab not a b
ffnothl ffnolth
ffd.tnfft
Rnilnol
1fI a
ab
Not
II no
Rnt
let flnig 2 have continuous wat rig 12
Further suppose
glt flatt yal 2t
 gf If n'HI fyy.lt
t
tff.lt
Fundamentaltheorem ofalgebra

Pn 2 Got 971 and anto

Ao Ai anf
72otcs.tpn20 0
BigO notation
an OBn if Int k7pm n No

smafnota.in an o pn if 4 to as new

General Error Formula

y flat

Inherent Truniation
error Error
Systemdependent 1190 pendent
can'tbemodified
canbe modified

Suppose
y Ffa nn f is cont havepartialderivatives

ytAy fastAM antAnnt

nltfm.AM fanAnn

i
Ay fm.AM fanAnn
Relative error
if is of higher dimension then use norm
affff
n

In offunctions Rel error f

ff.fm
ease

condition Def
if ftp.t nait
I consider finding Sol of three or more concurrent lines then mathematically we
need to find Sol of A'y
y An n
Numerically we can do it in the following
Choose a point on any one of the line
Findpoint onthe neighbour line which is projeltion ofthe previouspoint
Continue the
proless untill the point lice on every line
I

In consider hi flail data givenproperties ofdata we ran

interpolation

let plal Ao aint Ann an 0

net unknowns
from given data flail plai is onetype ofproperty we can define from this
we can get the polynomial An This process of defining the property is called
y
Interpolation

Polynomial in shifted form plat got anIn it t anCas

Thisform behave well numerially

than the originalpolynomialform
Newtonform shiftedpowerform with different shiftfactors
plat Got a n 41 an n a a ca t an a al ta en
complexity computation of patz
Plat aot.EE k a1 ai

Noofmultiplication 1 21 t n
nkf
multiplication or division takes moretimethan add or sub
 No
99
n 424 in n
If
Complexity n t ninth cn 0n't procedure
Remarks complexity can be brought down to 011

let n 3 pk a 6
if
set
Tf
1 as as
2 a as as 2 13
1 a outas 2 2

4 ad not at 2 51
2 ad Complexity n multiplications 2n additions
p
401mprocedure
Ag Input n
t.cn a

set an an
for
j n t to O
aj Aj ajtl 2 Citi

p21 ad
output ad p127
Suppose ni flail gil 0 data
Define plan dolocal an In a

aim
Ii Iii
III
 And Yr plan a

plat Effluil lifal Here

p
is called Lagrange interpolant

For n t local lift

11h Idm
Pilat
111g fool 4th find tho
thighs Innot
Weierstrass Approx theorem fful is cont on ab
so Pn sit YaffaPaul
f n flak
let Pn an polynomials of degree'n
if Panil an ail i Oton
Portal Anfal a

Q How to generate coefficients algorithmically

Pla P al n no n mic rful this polynomialsatisfies all points

from noto UK

flail plait Picki i O to k

since
y is of degree k
flail Pi Ini
Now Suppose flail P ni i O to k 1
I 9 fail flail i 1 to k

Define plat
114am IIa Palm

pno Picno f not

PINK 9kUk flan
 from 1 to K 1 plant
In acne 144in Rafael

nil nil
plae flue
I pful flat

Paint add Pick

III w

Suppose palm f no f nom n no f f no KK1 n no la he z

9 at f a f mn n m f m nu nm la net

I w.k.tp.int ffnoltflmtj.tk n no

flow m
flMIffs
Now to show floor flm t.me hk
ffno

P
I m
aim I w
t.im

p nil flui 1 0 to Ktl

pfulE Pa at n

coefficient of n in PaInt f no Mt
9k
weffilient of leadterm in RHS fwefficient ofleadterm in 9

weffilient ofleadterm in pic

twotfm Uti tho mic

Algo Input fail nil
for k 1 to n
i 0 to n k

f nitt nail flu

1
1 hi nail risk
Kirk Ui

output ao an

Ex N 1 23 4 5 6
flat 1 8 27 64 125 216
a Formulate divided diff table
b Givethe polynomial that fits the data values of 2,3 4

Pn a f no n no f no m n no n un i f non he

1st Vuln n not n Uk 1

Pn n f no
Effno hi Untal

kill a a

Given 2 we have to compute

p2
Ago Input f no nil i O to n set 0 f no an
ago
set U f no un for i to n

for k 0 to n 1 0 f ni un

0 12mild
f no Uk 12 k end
end
V p2

Recall Interpolation via polynomials in Newton form

Pnul not as n ro an n no nm n un 1
 we need to find ai s t flail Pnfail i O in
ai f no hi notation

Print
f no nil
j In a fin ni flui n.de
nikff
for n 4 no f no
mm
m a f noM M
FCRUs f no n
2 f n
f Mn n no no
flu n
flu n
flag
FEB 4 flu
a a
flu

I flat considered known

ant then Pnt n Palatt flao untilUnfullnun

In
anti In un t
here we're
estimating flat
Pnt nl Pnful f no Jun n un

for n ñ
Pneilal Paul If no an i Ulu u un

If no unit
f b a

Red Suppose f is continuously differentiable of order'k and no Uk are distinctpoint

in a b then 2f ab sit f no Mc f for any k

Proof set exin flat Petul

e not efml 0 when K C

By Rollei Theorem Fzffa.bls.tk 21 0 f 2 p2

1 21 F no 1 n no f no n P2 f nom
 Suppose f no nm
fly
em n flat PmMI

em nil 0 i Oto m

em no em m D 2 norm st email 0
em ml emuz 0 2n n n sit em 221 0

2 ab sit emm 27 0

0 emma Ff _pim a
D fÉ m f no um f no nm

Pnt ni Pnful b al

b at cf

Interpolation based on equispacedpoints

faib a no noth not2h notnh b

h Dna
flu flat nh
to app fn
nt a b

setting uny in in newton's interpolating formula

Pni Pn f no nnn n no n un

Pntiful flat Hat a b

fT Pn ñ f no unit i ry
I
no n 411
I flail intuit Iola n forsome ab
 Ifa infill I a

i
Iffal Priti 1
Pnl Of

Interpolation based on equispasedpoints

a no n both kn athh b h b

fi f not in a notsh
fful fs
Define Δ forwarddifference operator

fk fk I O
f fly Aite is 0

fic flat fi
Ife Atn Δ fact fill
ficiz fry fact fic
Define Efk Flat A E 1
shiftoperator
Property f no ni

k are equispated points In general flak naif

Proof when iO f no to
suppose the
eq is true for i m

then f no nmi flu time flu nm

 tho until n I.tn diff
4
Ps plnotsh ffno nu n
nil

II nosh no jul

IT Is.it
Notation sts ills
191

to
y.IE
Ps

If the interpolation is based on the Man

E
s
Ps P mish

it Elite
L
no to
Afo
be a
If
2 f A
Af
f 14
Af
f

for fi define f fit fi Aifi Aifil 21may Ei Exerlise

III.IT II uations may arise in normal interpolations where the error can
go higher for less change in points In hermite interpolation we
consider

the funi its derivative to be equal to the unknown funi its derivative
respeltively
ie flail plai i Oto n

nil p in 5 0 to m
For m 1 flail plait
1 0 to 12 2 conditions
flail p a

p to be a polynomial of degree
let us consider 2h11
Ai Bi are polynomials
P Aicalflail Bilal.fiuigy ofdegree 2ntl

from
Ailuit Si i plug ffnjl Bi nj finjl O.Aifni Sij
Bilnil
Si Aifnil
O Bilni ti j
W.k.tl a of degree n also satisfies lifaj Sij Ailni of degree 2nt1
so let us suppose Ailm faintbit
2nF

Bilal cintailfilm
for i j A nil 1 faintbillifa 1

ain bi 1 fi

similarly Billuil 1 cinitdi 2diluil I nil Ci diluil 1

2 rinitdil.li'fni Ii 1
Aiful faintbit 2 lifal.li n ai.li n
 i 0 2ainitbi limit ai a 2lifail O
ai Ili ni
bi 1 Aiki 2milifni
Bilal O Cini dit e nil 0

limit di 0 i e dis ni

substituting allthese in eq's we get plal aintbil.eicnffniltflintdil.lifnlfln

plat zeitai 1 2nihilai dial fail

In nil.lial.flni
plat liarflail n nil 21cnil 1 1

1m Given the following values of flat film estimatethe values of ff05

fin f 0.5
a flat
I 1 5
O l l
l 3 7

finalans 2nF n nil

 i i ii continue suntan
We divide the polynomial into piecewise polynomials to make surethat each
pietewice polynomial is similar to the given fun

plat P.fm nffa.ci

eEh P2 n nt ink

a d d k in O
Pial a a a

In this way we can find the errors in specific intervals make changes

asiordingly
whenthere is a large spike in a short interval polynomial interpolation gives
a polynomial of degree in all intervals dividing them into
high
intervals spelifi

Tolynomials give higher degree polynomial in the spike region regular polynomial in
agg
Sielewise Linear Interpolation

Given

hi
i i h h in
P.nl find
file ffniltntfno.nl
14h
Palm tnta.nl
I find
III tea

Pi.int ffai.tt J.ffnil.tntCai i.ni

I I
 Plat Pinful EININflail

I t.am and
1

I Iiiiiit
am.li
T.EE
find i
I m m Aml

If tend
I am

NoInt ntlno.nl
Itm Nila 11h ntcno.nl

ntcm.nl

Nn Anthmian
1 1 1 Nim 1

Ni attain.mil
Tm hi hi nitt In1 an
ntfni.mil

Error elal
p al flat at a b

nf no nil eful flat P n

nt hi ini elal flat Pi tn

i
If
a ni i fanil ni i

f ni r nil M
 rewise Quadratic Interpolation we dividethe polynomial into quadratil polynomials
over intervals ni nitz

im
natant in
find to h ne as he thetinyunam

Palnl Piz Ni
pietewise
quadratil

Bivariate Interpolation Suppose we have a two variable polynomial 2 flu y we

are given data ni.si

ffg.li j
8I Pm.ntni47 Effffi.j le ilnlli.jlyl
e.im C
if
t.int Iq
Tpin a Pierewise Smooth fun's
F is a spline of order m

1 Flail flail ti Oto n

a F'm is cont in

consider data nitrd Yier

d
Cybenko f is continuous on 0,1
Given 70 JNE It WIE R bit R

affa flat win bill of

howto compute values

flat fwi.mg

is a non linear cont funi satisfying 11 1 too

0 tx 0

Ex alt Sigmoidalfun
t
Loss
9 lwjy.tk
Hunt
a of Wi bi
we calculate min loss Wibil WitRd bifR
givesthe optimal values

f 0,1 Rm is continuous Hiven 30 K Ai bi s t Ai E 121Mt bi

i to K I
s.tn n
n Ain tbi 6 is a non linear
sigmoid funi
n A n bn

n A nl bic s.tl ffal nfkl at Ai aretriangularmatriles

 ni.it n n

I ni y
i

i mtkt nitrite
ia
1ˢᵗlayer 1k4thlayer
loss 11y ni 112
1,1
AK I bk
AMY I
Yi Ak I be

Q n 0 I 1 3
ffa 1 2 33 244 Fit a quadratilspline with f o O

nscendental Equations
Cybenko
Geometry ofDL
f IRIE Suppose f changes sign
y flat
consider

in a b i.e flat F b 20 we want to find no in faib

topmfutational

st fino O
L
Aside If f is cont on ab It Me Megyn
f

3 aib sit f2 8
Now here we set 8 0
Biseltion Method

take no a m b
take v
LY f af
if f r flnok 0
No No M 6

irriterion
ii ii i
to stop the process i un i un 28 3
i
m

disadvantage
if 0 flani a E maybe expensive numerially
Sffergen

Algo no a M b Algo 1 No a Msb

while 1m notes for r 1121
Lifton
hotf w 4
if f no tfr c D
if f w flan11 0
No No M 8
else if ffnol.fr 0 else
ff.fiwffnn ilso
No V M M
Usfreak
break end
end while bFIM

Regula Falsi Method consider funi f

1 b
y FM n b
I
n Wi D
y la f
flbl l w b
 Ni a.lftblltb.lflal
afffj.fi F1b7 1flall

Now we consider the line from w if wil to fb.fm

repeat the process until Wn 2007 8

Problems Find the root of m n t o

a fful
b flat wsion
1 flat 0.3 0.5 n 1

0.3 Oen 0.5

flat m n 7 flit l l l 1
f2 8 21 5
since flil f 2 20
wi

I
fail I I 2

a fful ffol 1 ffil

since flol f 1 CO
W 0 1

flail I
 we t

f a

us

flag N diverges i flat is not cont

in 10117
b cosion has 3 roots in fo I
them is returned
anyone of

i'all the iterativepoints areone

029 y side of the root
The regular falsimethod results
may converge veryslowly or may show wrong
if the fun is convex So we canTemodified regular falsi method for fast
convergence

Algo Input a b f b flbl

sit flatfenko
Ao a b b F flao f Hbo
y
Wn rapidly

iii I a

if flan Fluent 0
bn I Write Ant An G ffbn.nl
 if fluent fluent 0
set F Flz

but bn anti Wnt F f anti

if f un fluent 50

set G 4 2

Then f has a root in anti bat

Seiant Method

while n 1,2 IfCantill E

How isthis different from Regular FalsiMethod I

NIRaphsonmethod check convexnotes

while n 1,2 flantil f

una un
FYI
let me Roth
fluil flnoth
f no t n f no Tangent approx

Eq of tangent at no f not f not n n

y
n m O
y
O f no f

uol. n m noth no
Iff
 anti an
ff what is the guarantee that this sequente converges

In Newton Raphson method we start with a guess

continue the above proless
The convergenicmay depend on initialguess I
the that arise in this method are ftp
If
questions

What is the guarantee that an 3

flan
Speed of convergence
Them Suppose suffiliently nite funi f is such that flu O is equivalent to n 01
where is differentiable If E is rootof f f all nF I where I isthe
interval containing 3 thenthe sequence an defined via
un un t n 1,2
converges to I
104m71 kc1
Proof 31 0 2 012

nn Inni
n 3 Ann1 0131 0in fun z a

fun 31 n funi 21
K an1 31
k nnn 21

114
K

0 a new

061
suppose n
Iff glut I
ftp.f fff
 14141121 4 at ff I If't Ant I
ntI

Then if not I an unt 2

21 answers question we need to choose
of s t above condition is
satisfied then Un 3

ex flat Fr n r

f In ron

P
Def An iterative method is said to be oforder if un 21 un t 3
p
an 3 with order ofconvergence beingp
Suppose for un 3 an Z En
anti nn
ff Ent th

I
a

cnn.tn

fan en
i iitiii i
fni.ci Into a

f ti
Ent En Ent
Iff
1 en
fifth
fat f Ed Iff often

fail tentM
anti 31 un 21 M
 Aircrew fuk 014k17

ni non n

ni 1 fasterthan nut proless mention

page
escenario
t.ua ffn.gl so Assumethat there is a point common to both
gln.gl o
ffn.gl 0 n Flury
Fit have continuous partialderivatives
ginny 0
y gang
Further if Fn Gate1 Fy try af over I then for nogolf I

Uk F dec1 Yk i 12 1 2
n
9K M UK 4 1 y
ino
unyn 21,321

fii.fi
5 0137 137 0
ii

Z is fixedpoint of y 1M
Q Find real root of fful u n 1

Q2 Findroot of 2n cosn 3 correltto 3decimalplates

Qs Perform 3 iterations of NewtonRaphsonMethod to obtain approximate value of 1177
Dealing withpolynomials Pn do ant and an 01
We need to find St Pn31 0
DistortedRule Noof positive roots of pn Xp Noof signchanges in
Ao as an
V Isay
then Np V is an even integer
Noofnegative roots of Pn no of sign change in coeffilients ofPntm

It f in circle ofradius minffo.tn

I f
if a root
n then p has

If It then all theroots of Pn lie in cirile of radius r around 0

8
7.8 1

Inverse interpolation

Solving Matrix Equations Ah we assume A is square invertible

y
There are 2 classes of methods Direst Methods Guass Elimination myth
Iterative Methods
MHYomposition
Grass Siedel Jacobi more move
Qualitative aspects of soli of
g An
at an b A
b

symmi
 data t.sn in

suppose take another

system AT S is

i b
Hunsman

if Changesin
input can impair
the cot more

MatrixNore F R IR thetalgoshould be
setof all stable to provide
nxn matinee over IR
Correllresults
If F satisfies F A B FA FIBI
Flaal 1 1FCA αfR ACRm
FA 0 A O
FABI FAI FCB A B RMM then F is calledmatrixnorm denoted

by 1111
2
Exile if IAll Iai Frobenius norm

in 11AM
1 Pt 01

I
1 4vector norms
Ime called subordinatenorm

Result 11All Max 11411 1111211 Kinll

Ci ithcot of A
nil
nut
 IAllo 4trill
may 11h11

ri ithrow of A
11Ally
Flat flat maxsingular
valueof A
maxtigenvalue of Ata

Result An b An b A A18A b b ish

then iA 9th litousain

KAl
Landy on in a
hinge inn
so

Ex M 242 3
2M 3.9994 5.999 Sol is 1,1

Perturbed
change in sot is huge evenfor
µ 2n 3
small change of input
2m 3.99912 6 501 s 3101 so nature ofSol is needto
be discussed before algo

Prefaboverult An D
AISA n Sn biSbl

ASnt SAN SA.sn Sb

late
A It ASA 8h Sb SAn n willbeproved
Suppose 11H11118AM 1 then

Sn HASATA Sb San
invertible

Ifsa
A
I'smany Fill

fi ali sa
 it.fi ii liait it
b An Hall all
HI
11611

a.fiitnl
KAI IAll11A I

I int
1
KAI Cixi d
nx.fi
kA 11 1T
e it 0111841 KAI All1A'll
Kpfa
lp
HAllp.UA

Landginson
oh in a
flange inn

when
p2 K A IAll IA'll XFAT.A.xminAT.AT
B A
then xma.is
tat
k Al I A is unitary

Pre treatment Suppose It Flu 0 where u O

u Fu
11411 all lull 4 0 contradiction

i It F exists
 IFI IFI It F F I It FIF

111ftFI'll 11 I It FIF I
11111 11 FTHHFII

Popribadmatid U a a Un f R

A
4 I I I un Vandarmondematrix

Jan
ai un
If U 1 7 34 5
bond Al 2.6170 104

Hisbert Matrix A if n 10 Condal 1.6025 10

aij HT mn

let Alt
conditioning of eigenvalue problem
p nun

if f O alleigen value of A O
f 18 n 40 eigen values have magnitude 0.1

net Altl

an I
E O eigenvalues 1
f 10 eigenvalues goaway from 1
 Use Regular Falsi method to find the smallest ve root of n t o correlt to
3 decimal places
Q2 Use Newton Raphson method to find a root of Ulogon 1.2 0 correltto

Q3 Use the bisention method to find a root of equation fin n 4 n n

in the interval 1,9 and Ian
you predilt at what iteration the
method converge

QI Determine the no of iterations for the biseltion method necessary to solve

flat m dm 10 0 with accuracy 153 using a 1 b 3

3b we know that Pn Pl E b a

where E error or tolerante

N no of iterations

103 103
13 1 2
2

INI
Therm PAP
Suppose Ann is sit Diagonal x x2 Xn

1Diag Xn I
Max til
x x

Then in speltrum AtSA 2 12 it 118All k P

19 p if k is verysmall then
In will be closer to Xi
ie eigenvalues of A

If k PI 1 then P is unitary
I µ 2 12 til 118AM
 ii.am.am
92hi 922 2 923h be
93m 92h2 933ns b
Suppose Aii 0 i 1 2,3

M 9 as t
Aug 1 notBust8

2 m
9 am pint
s 83
9 4 n 84 13 2

be the guessfor sot of

n n n
y Au
Suppose

n α nl t βn 2 nl t
0 nl t α β.nl d
n α n
βn d α nl 0nl
β
n's di

a α n pen 8 α ni pen 0 nl 8

n n 8
p
T T
l kt
what istheguaranteethatthis will converge
n un t n I
to the Sol
Suppose A is manipulated sit A I U L

L.nl t
then 2kt U.nl
f
I L a U.nl I
Suppose L is sit I i is invertible then n I n t
Info
 a
kt
c mlk f
n 1n 8
kt k ll
a n a nlkl.pe

C n
l n C n n

1in n 11 11111

1 ÉL
if Hillel 11111 t 0 at kt N
10 we should make sure that we can write A It Lt U II 1 is invertible
111 11 4011 1

JawbasmethI An
y
is Sol
y II Ala
n when n

we ran take n kt I A n
yt
n sot when I I All Cl
simultaneously all components gets updated

2519

y
An y ain i i in A
9
We will represent y air at concurrentlines all passingthroughthe sot
If we follow the method shown in figure 9 ai n

sol of An
y
 z projertion of 3 onto

gi
I a
Fw
1 0 is
3 ᵗY scope
Yitalk
Suppose to is sit 2101
tou lies on a n
y
tou
fan 11 KanmarsMethod

a 3101 to a a
y
to an 11am
3 9

i n of a
1 projertion on y aim

n n

Alspanut
I
mix
Hemmed any

Ffu OfUtR
Mtn My tn.tl
either upper or lowertriangular

t
H IHall 11h11
I
21 H is symmetri
A Hu U
1H 21 v.v v spank
ᵗ
Householder
reflation of a
want theplane 24,1T
a
26101
Ha In 2fa.at u u
fyCuut a

Hr v v ve spank

Suppose 21.2 R are s t a 121 24 1121112 Hell

then 21 22 H22 2

21 21

H2 H 2 H
II
all 2122
constant

Hu Han
a Xu HC II
21 22 21 21 112111 112211 D
t
21 22 2,122 2,12 f spanka
i 2 2

22
Finally Hz 2121 2

Theorem's suppose a n to ER set

y Hallie i set u n y

then H I 2,4
Hn y linthe

011
Read Difmethod An b
MA AT
11 IT multiplier
maps
 via banksubstitution n can be obtained A at b bi
Alg for bank substitution kn bilann
For K n 1 1
Uk bi Efficient
output Mn 501

In householder method we take a ay 1ˢᵗcotofA

Haillie m n
y a 1191124
y
Him

HA Hia H G Him

ai y Hailly n

iffj
obtain
man

Hittin Hi He
1

I
f
Q A find Hi He t
go
Hattia
f
 a

m ar t Hallie
Ifc

Guale Elimination Method

Input A b

ii iii o
notexist
Sol does

partial pivoting
swapkthrow with ethrow
where I argamax art
end

Hifi
Kalin
For
j

p iqiiygy
else barksubstitution

end
 LV decomposition method

Therm Suppose Ann is a matrix s t All kill in are

matrix U sit
non singular Then a lowertriangularmatrix L upper triangular

A LU Further if Lii I i then1401

pair is unique

LET MT
u Decomposition A LV

An b Lun b we have to solve

Proof for above theorem At 14ᵗʰstep

1m
Iii
O

Lk AK
I DK
H
At end A U
A Tv LU

Univ Suppose A 40 1202

iii vi
Ii 7
Lii 1 i
Cholesky Decomposition

1 If Ann is positive
definite symmetri matrix then a lowertriangular matrix L s t

A Lt holds Tanso an
i e all eigenvalues of ASO

41 Further if Li 0 faitorization is unique

If A LT holds then A is symmetri
UTAN KILLIN

lini Tn ILinll 20

Consider the call An where A 1min mon

y
then no ns.t An wolds
y
So we try to approximate the Sol byfinding myan An bll

fin An bl12
An b An bl

tie
21 0

1,1 O

IRM
Consider
Y IR
y Wn

if y An A
11
 utan nT AtATI
y
An
An b An b Natan btan stab btb
y
rankAl n

11 0 I ATAT'Atb

Rapprox ATAT'ATb

Atb where At ATATAT pseudo inverse of A

Considerthe case An where A 1min man rank A M

y
then infinitely many solutions exist
several different criteria on sol we get several differentsoles
By imposing
if it is sole then y A ñ Xy
one criterion min distance condition min link sit y An
1141124 x y Ant
Mfr
flat Hall XtlyAnl
If O n ATAATT'b

Another criterion c n i nito

min Ifn sit An
y