Lecture 21 Let A Along FCA E Ak A X MAI CA KJ M Moje Mao A ...

2
CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 Lecture 21 (last time) Uses of matrix functions for graphs Uses of generalized eigenvalues and how to compute generalize eigenvalues. A new class of methods to solve Ax=b called “Krylov methods” CG, MINRES, SYMMLQ, GMRES, LSQR These methods search for the best solution over a subspace and they will always be better than Richardson, Steepest Descent, etc. Then we hyper-optimize the implementations, which really dier between symemtric and non-symmetric. This will give us new algs for eigenvalues and eigenvectors too! These methods have many interpretations; - polynomials - subspaces and optimization - conjugate directions (e.g. CG) Mahixfmetions Let A beak adyacag CA KJ ,j= # of Along walk probably M - FCA ) Moje E ¥ Ak to A t Genteel X 's Mao MAI Ax=yB× A±=XFFIx FCA ) - means apply Matrix af a grant . isnt informative . =I+×A+£AI£Az , Linear Disarm .hn Auto . 7 I Bx fad E the G. CUE ) Walks between ianlj . so people befit M= exp ca ) / I one statures point . Eigenvalues . of length K . Supra are hee dad f- ' A ,×=ypT± f. ÷ : ÷÷÷÷÷÷÷ . ÷i÷ : man f÷÷÷÷*÷÷÷÷÷÷÷.f± : Lets use these for gw.es . way bgn Kate rebound . a 't th ' " easy ? Studying graphs t VII. 43.41 a weighs a , a gram F- ' AFTF = dy Network ! E- C Cl , 2%3 ) , can , Cd4 ) " convoy " AE XI are Let Be pptfe the 4. y , , cz.gg - MY not # it a sthonary points ' cheeky femur ( s net smell Enough .

Transcript of Lecture 21 Let A Along FCA E Ak A X MAI CA KJ M Moje Mao A ...

Page 1: Lecture 21 Let A Along FCA E Ak A X MAI CA KJ M Moje Mao A ...

CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019

Lecture 21 (last time)Uses of matrix functions for graphsUses of generalized eigenvalues andhow to compute generalize eigenvalues.A new class of methods to solve Ax=bcalled “Krylov methods”CG, MINRES, SYMMLQ, GMRES, LSQRThese methods search for the bestsolution over a subspace and theywill always be better thanRichardson, Steepest Descent, etc.Then we hyper-optimize theimplementations, which really differbetween symemtric and non-symmetric.This will give us new algs for eigenvalues and eigenvectors too!These methods have many interpretations;- polynomials- subspaces and optimization- conjugate directions (e.g. CG)

Mahixfmetions

Let A beak adyacag CAKJ ,j=# of

-

Along walk probably M-

- FCA ) Moje E ¥ Ak to A t Genteel X 's. Mao MAI

→Ax=yB× A±=XFFIx

FCA ) - means apply Matrix af a

grant . isnt informative .

=I+×A+£AI£Az,

Linear Disarm .hn Auto . 7I Bx

fad E the G. CUE ) Walks between ianlj . so people befit M= exp ca ) / I

one statures point.

Eigenvalues .

. of length K.

⇐ Supra arehee dad

f-'

A ,×=ypT±

f.

÷: ÷÷÷÷÷÷÷. ÷i÷:man

f÷÷÷÷*÷÷÷÷÷÷÷.f±::;¥Lets use

these for

gw.es. way bgn

Kate rebound . a 't ⇒ th '"

easy?Studying graphs t VII. 43.41 a weighs a ,

a

gram .

F-'

AFTF =

dyNetwork ! E- C Cl, 2%3 )

,can ,

Cd4 )"

convoy" AE XI are Let Be pptfe the

4. y ,, cz.gg

-

⑧ MY not

#it a

sthonary points'

cheeky femur(

s net smell Enough .

Page 2: Lecture 21 Let A Along FCA E Ak A X MAI CA KJ M Moje Mao A ...

CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019 CS 515 - Purdue University - Computer Science - David F. Gleich - 2019

Method to sohe A 't :{Given Ay I

wecovh try We one

goingto see an > A "=

I+c±→ , # a , ,+.

Observe Quiz : Soln :

Consider the promNot that

First idea to get a What he here how ?CAgain ) it Aisle

.ge , sparse AtA±=aT£amaub class 4 meth

Consider he lghzl, (IAJk=§gCjgc_Ajb Ihae XI"= pat ) b- better method !

> We car ↳ he £

Hoetger www.T?heAgFn.

Bean ass↳ MM . cos

⇒ New "

F.method !

E"=[s⇐→s⇐ a ,

. .⇐a ,

,kg±31

'' → fm Tt

poxs=Ea→÷f⇒. \pbua

.

the soso.me

* " hmm

.VN#bbsid

. £⇒mfeEkmdan,

is.ie?itmfem:eaeEe;fEiEE.;YInf.YYyaa:nf"

¥⇐" "

fast :*::*:* HEFEI'.EE#t:gefoEYiii9.aitfieioEEYaFfxIi:YEIEIIse.ii:ieasHiEEieYEIHEiaEEEEafEiEeIHtais.fe?.;wm*a

x.= a.

,

§a

at

Obsewh 2Ex : k=2 Istheeaoednc

y=[ b. As, ,Azkf]<

GMRBS.

thestory Class

.

× "4=C b

,Ab

, -

,Atyjc

C k€33 )×

" ' ist # HE taaiskdthk ) =

span c bag .

Str -

=b+b - At +

c±→a+* , ,£ Aksj [ birth - ,A 's ) c- is B

£ sore c-.

=3b - 3A§ + ay =) dim Ka ( a. b) act ,abetter

approx .

to x. ?⇒ Y= BE

\ I I 4 Mid 11 b- ABELE-

[ 2